1 // SPDX-License-Identifier: GPL-2.0
3 * Copyright (C) 2007 Oracle. All rights reserved.
7 #include <linux/blkdev.h>
8 #include <linux/radix-tree.h>
9 #include <linux/writeback.h>
10 #include <linux/workqueue.h>
11 #include <linux/kthread.h>
12 #include <linux/slab.h>
13 #include <linux/migrate.h>
14 #include <linux/ratelimit.h>
15 #include <linux/uuid.h>
16 #include <linux/semaphore.h>
17 #include <linux/error-injection.h>
18 #include <linux/crc32c.h>
19 #include <linux/sched/mm.h>
20 #include <asm/unaligned.h>
21 #include <crypto/hash.h>
24 #include "transaction.h"
25 #include "btrfs_inode.h"
27 #include "print-tree.h"
30 #include "free-space-cache.h"
31 #include "free-space-tree.h"
32 #include "inode-map.h"
33 #include "check-integrity.h"
34 #include "rcu-string.h"
35 #include "dev-replace.h"
39 #include "compression.h"
40 #include "tree-checker.h"
41 #include "ref-verify.h"
42 #include "block-group.h"
44 #include "space-info.h"
46 #define BTRFS_SUPER_FLAG_SUPP (BTRFS_HEADER_FLAG_WRITTEN |\
47 BTRFS_HEADER_FLAG_RELOC |\
48 BTRFS_SUPER_FLAG_ERROR |\
49 BTRFS_SUPER_FLAG_SEEDING |\
50 BTRFS_SUPER_FLAG_METADUMP |\
51 BTRFS_SUPER_FLAG_METADUMP_V2)
53 static void end_workqueue_fn(struct btrfs_work *work);
54 static void btrfs_destroy_ordered_extents(struct btrfs_root *root);
55 static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
56 struct btrfs_fs_info *fs_info);
57 static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root);
58 static int btrfs_destroy_marked_extents(struct btrfs_fs_info *fs_info,
59 struct extent_io_tree *dirty_pages,
61 static int btrfs_destroy_pinned_extent(struct btrfs_fs_info *fs_info,
62 struct extent_io_tree *pinned_extents);
63 static int btrfs_cleanup_transaction(struct btrfs_fs_info *fs_info);
64 static void btrfs_error_commit_super(struct btrfs_fs_info *fs_info);
67 * btrfs_end_io_wq structs are used to do processing in task context when an IO
68 * is complete. This is used during reads to verify checksums, and it is used
69 * by writes to insert metadata for new file extents after IO is complete.
71 struct btrfs_end_io_wq {
75 struct btrfs_fs_info *info;
77 enum btrfs_wq_endio_type metadata;
78 struct btrfs_work work;
81 static struct kmem_cache *btrfs_end_io_wq_cache;
83 int __init btrfs_end_io_wq_init(void)
85 btrfs_end_io_wq_cache = kmem_cache_create("btrfs_end_io_wq",
86 sizeof(struct btrfs_end_io_wq),
90 if (!btrfs_end_io_wq_cache)
95 void __cold btrfs_end_io_wq_exit(void)
97 kmem_cache_destroy(btrfs_end_io_wq_cache);
100 static void btrfs_free_csum_hash(struct btrfs_fs_info *fs_info)
102 if (fs_info->csum_shash)
103 crypto_free_shash(fs_info->csum_shash);
107 * async submit bios are used to offload expensive checksumming
108 * onto the worker threads. They checksum file and metadata bios
109 * just before they are sent down the IO stack.
111 struct async_submit_bio {
114 extent_submit_bio_start_t *submit_bio_start;
117 * bio_offset is optional, can be used if the pages in the bio
118 * can't tell us where in the file the bio should go
121 struct btrfs_work work;
126 * Lockdep class keys for extent_buffer->lock's in this root. For a given
127 * eb, the lockdep key is determined by the btrfs_root it belongs to and
128 * the level the eb occupies in the tree.
130 * Different roots are used for different purposes and may nest inside each
131 * other and they require separate keysets. As lockdep keys should be
132 * static, assign keysets according to the purpose of the root as indicated
133 * by btrfs_root->root_key.objectid. This ensures that all special purpose
134 * roots have separate keysets.
136 * Lock-nesting across peer nodes is always done with the immediate parent
137 * node locked thus preventing deadlock. As lockdep doesn't know this, use
138 * subclass to avoid triggering lockdep warning in such cases.
140 * The key is set by the readpage_end_io_hook after the buffer has passed
141 * csum validation but before the pages are unlocked. It is also set by
142 * btrfs_init_new_buffer on freshly allocated blocks.
144 * We also add a check to make sure the highest level of the tree is the
145 * same as our lockdep setup here. If BTRFS_MAX_LEVEL changes, this code
146 * needs update as well.
148 #ifdef CONFIG_DEBUG_LOCK_ALLOC
149 # if BTRFS_MAX_LEVEL != 8
153 #define DEFINE_LEVEL(stem, level) \
154 .names[level] = "btrfs-" stem "-0" #level,
156 #define DEFINE_NAME(stem) \
157 DEFINE_LEVEL(stem, 0) \
158 DEFINE_LEVEL(stem, 1) \
159 DEFINE_LEVEL(stem, 2) \
160 DEFINE_LEVEL(stem, 3) \
161 DEFINE_LEVEL(stem, 4) \
162 DEFINE_LEVEL(stem, 5) \
163 DEFINE_LEVEL(stem, 6) \
164 DEFINE_LEVEL(stem, 7)
166 static struct btrfs_lockdep_keyset {
167 u64 id; /* root objectid */
168 /* Longest entry: btrfs-free-space-00 */
169 char names[BTRFS_MAX_LEVEL][20];
170 struct lock_class_key keys[BTRFS_MAX_LEVEL];
171 } btrfs_lockdep_keysets[] = {
172 { .id = BTRFS_ROOT_TREE_OBJECTID, DEFINE_NAME("root") },
173 { .id = BTRFS_EXTENT_TREE_OBJECTID, DEFINE_NAME("extent") },
174 { .id = BTRFS_CHUNK_TREE_OBJECTID, DEFINE_NAME("chunk") },
175 { .id = BTRFS_DEV_TREE_OBJECTID, DEFINE_NAME("dev") },
176 { .id = BTRFS_FS_TREE_OBJECTID, DEFINE_NAME("fs") },
177 { .id = BTRFS_CSUM_TREE_OBJECTID, DEFINE_NAME("csum") },
178 { .id = BTRFS_QUOTA_TREE_OBJECTID, DEFINE_NAME("quota") },
179 { .id = BTRFS_TREE_LOG_OBJECTID, DEFINE_NAME("log") },
180 { .id = BTRFS_TREE_RELOC_OBJECTID, DEFINE_NAME("treloc") },
181 { .id = BTRFS_DATA_RELOC_TREE_OBJECTID, DEFINE_NAME("dreloc") },
182 { .id = BTRFS_UUID_TREE_OBJECTID, DEFINE_NAME("uuid") },
183 { .id = BTRFS_FREE_SPACE_TREE_OBJECTID, DEFINE_NAME("free-space") },
184 { .id = 0, DEFINE_NAME("tree") },
190 void btrfs_set_buffer_lockdep_class(u64 objectid, struct extent_buffer *eb,
193 struct btrfs_lockdep_keyset *ks;
195 BUG_ON(level >= ARRAY_SIZE(ks->keys));
197 /* find the matching keyset, id 0 is the default entry */
198 for (ks = btrfs_lockdep_keysets; ks->id; ks++)
199 if (ks->id == objectid)
202 lockdep_set_class_and_name(&eb->lock,
203 &ks->keys[level], ks->names[level]);
209 * Compute the csum of a btree block and store the result to provided buffer.
211 static void csum_tree_block(struct extent_buffer *buf, u8 *result)
213 struct btrfs_fs_info *fs_info = buf->fs_info;
214 const int num_pages = fs_info->nodesize >> PAGE_SHIFT;
215 SHASH_DESC_ON_STACK(shash, fs_info->csum_shash);
219 shash->tfm = fs_info->csum_shash;
220 crypto_shash_init(shash);
221 kaddr = page_address(buf->pages[0]);
222 crypto_shash_update(shash, kaddr + BTRFS_CSUM_SIZE,
223 PAGE_SIZE - BTRFS_CSUM_SIZE);
225 for (i = 1; i < num_pages; i++) {
226 kaddr = page_address(buf->pages[i]);
227 crypto_shash_update(shash, kaddr, PAGE_SIZE);
229 memset(result, 0, BTRFS_CSUM_SIZE);
230 crypto_shash_final(shash, result);
234 * we can't consider a given block up to date unless the transid of the
235 * block matches the transid in the parent node's pointer. This is how we
236 * detect blocks that either didn't get written at all or got written
237 * in the wrong place.
239 static int verify_parent_transid(struct extent_io_tree *io_tree,
240 struct extent_buffer *eb, u64 parent_transid,
243 struct extent_state *cached_state = NULL;
245 bool need_lock = (current->journal_info == BTRFS_SEND_TRANS_STUB);
247 if (!parent_transid || btrfs_header_generation(eb) == parent_transid)
254 btrfs_tree_read_lock(eb);
255 btrfs_set_lock_blocking_read(eb);
258 lock_extent_bits(io_tree, eb->start, eb->start + eb->len - 1,
260 if (extent_buffer_uptodate(eb) &&
261 btrfs_header_generation(eb) == parent_transid) {
265 btrfs_err_rl(eb->fs_info,
266 "parent transid verify failed on %llu wanted %llu found %llu",
268 parent_transid, btrfs_header_generation(eb));
272 * Things reading via commit roots that don't have normal protection,
273 * like send, can have a really old block in cache that may point at a
274 * block that has been freed and re-allocated. So don't clear uptodate
275 * if we find an eb that is under IO (dirty/writeback) because we could
276 * end up reading in the stale data and then writing it back out and
277 * making everybody very sad.
279 if (!extent_buffer_under_io(eb))
280 clear_extent_buffer_uptodate(eb);
282 unlock_extent_cached(io_tree, eb->start, eb->start + eb->len - 1,
285 btrfs_tree_read_unlock_blocking(eb);
289 static bool btrfs_supported_super_csum(u16 csum_type)
292 case BTRFS_CSUM_TYPE_CRC32:
293 case BTRFS_CSUM_TYPE_XXHASH:
294 case BTRFS_CSUM_TYPE_SHA256:
295 case BTRFS_CSUM_TYPE_BLAKE2:
303 * Return 0 if the superblock checksum type matches the checksum value of that
304 * algorithm. Pass the raw disk superblock data.
306 static int btrfs_check_super_csum(struct btrfs_fs_info *fs_info,
309 struct btrfs_super_block *disk_sb =
310 (struct btrfs_super_block *)raw_disk_sb;
311 char result[BTRFS_CSUM_SIZE];
312 SHASH_DESC_ON_STACK(shash, fs_info->csum_shash);
314 shash->tfm = fs_info->csum_shash;
317 * The super_block structure does not span the whole
318 * BTRFS_SUPER_INFO_SIZE range, we expect that the unused space is
319 * filled with zeros and is included in the checksum.
321 crypto_shash_digest(shash, raw_disk_sb + BTRFS_CSUM_SIZE,
322 BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE, result);
324 if (memcmp(disk_sb->csum, result, fs_info->csum_size))
330 int btrfs_verify_level_key(struct extent_buffer *eb, int level,
331 struct btrfs_key *first_key, u64 parent_transid)
333 struct btrfs_fs_info *fs_info = eb->fs_info;
335 struct btrfs_key found_key;
338 found_level = btrfs_header_level(eb);
339 if (found_level != level) {
340 WARN(IS_ENABLED(CONFIG_BTRFS_DEBUG),
341 KERN_ERR "BTRFS: tree level check failed\n");
343 "tree level mismatch detected, bytenr=%llu level expected=%u has=%u",
344 eb->start, level, found_level);
352 * For live tree block (new tree blocks in current transaction),
353 * we need proper lock context to avoid race, which is impossible here.
354 * So we only checks tree blocks which is read from disk, whose
355 * generation <= fs_info->last_trans_committed.
357 if (btrfs_header_generation(eb) > fs_info->last_trans_committed)
360 /* We have @first_key, so this @eb must have at least one item */
361 if (btrfs_header_nritems(eb) == 0) {
363 "invalid tree nritems, bytenr=%llu nritems=0 expect >0",
365 WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG));
370 btrfs_node_key_to_cpu(eb, &found_key, 0);
372 btrfs_item_key_to_cpu(eb, &found_key, 0);
373 ret = btrfs_comp_cpu_keys(first_key, &found_key);
376 WARN(IS_ENABLED(CONFIG_BTRFS_DEBUG),
377 KERN_ERR "BTRFS: tree first key check failed\n");
379 "tree first key mismatch detected, bytenr=%llu parent_transid=%llu key expected=(%llu,%u,%llu) has=(%llu,%u,%llu)",
380 eb->start, parent_transid, first_key->objectid,
381 first_key->type, first_key->offset,
382 found_key.objectid, found_key.type,
389 * helper to read a given tree block, doing retries as required when
390 * the checksums don't match and we have alternate mirrors to try.
392 * @parent_transid: expected transid, skip check if 0
393 * @level: expected level, mandatory check
394 * @first_key: expected key of first slot, skip check if NULL
396 static int btree_read_extent_buffer_pages(struct extent_buffer *eb,
397 u64 parent_transid, int level,
398 struct btrfs_key *first_key)
400 struct btrfs_fs_info *fs_info = eb->fs_info;
401 struct extent_io_tree *io_tree;
406 int failed_mirror = 0;
408 io_tree = &BTRFS_I(fs_info->btree_inode)->io_tree;
410 clear_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
411 ret = read_extent_buffer_pages(eb, WAIT_COMPLETE, mirror_num);
413 if (verify_parent_transid(io_tree, eb,
416 else if (btrfs_verify_level_key(eb, level,
417 first_key, parent_transid))
423 num_copies = btrfs_num_copies(fs_info,
428 if (!failed_mirror) {
430 failed_mirror = eb->read_mirror;
434 if (mirror_num == failed_mirror)
437 if (mirror_num > num_copies)
441 if (failed && !ret && failed_mirror)
442 btrfs_repair_eb_io_failure(eb, failed_mirror);
448 * checksum a dirty tree block before IO. This has extra checks to make sure
449 * we only fill in the checksum field in the first page of a multi-page block
452 static int csum_dirty_buffer(struct btrfs_fs_info *fs_info, struct page *page)
454 u64 start = page_offset(page);
456 u8 result[BTRFS_CSUM_SIZE];
457 const u16 csum_size = fs_info->csum_size;
458 struct extent_buffer *eb;
461 eb = (struct extent_buffer *)page->private;
462 if (page != eb->pages[0])
465 found_start = btrfs_header_bytenr(eb);
467 * Please do not consolidate these warnings into a single if.
468 * It is useful to know what went wrong.
470 if (WARN_ON(found_start != start))
472 if (WARN_ON(!PageUptodate(page)))
475 ASSERT(memcmp_extent_buffer(eb, fs_info->fs_devices->metadata_uuid,
476 offsetof(struct btrfs_header, fsid),
477 BTRFS_FSID_SIZE) == 0);
479 csum_tree_block(eb, result);
481 if (btrfs_header_level(eb))
482 ret = btrfs_check_node(eb);
484 ret = btrfs_check_leaf_full(eb);
487 btrfs_print_tree(eb, 0);
489 "block=%llu write time tree block corruption detected",
491 WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG));
494 write_extent_buffer(eb, result, 0, csum_size);
499 static int check_tree_block_fsid(struct extent_buffer *eb)
501 struct btrfs_fs_info *fs_info = eb->fs_info;
502 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices, *seed_devs;
503 u8 fsid[BTRFS_FSID_SIZE];
506 read_extent_buffer(eb, fsid, offsetof(struct btrfs_header, fsid),
509 * Checking the incompat flag is only valid for the current fs. For
510 * seed devices it's forbidden to have their uuid changed so reading
511 * ->fsid in this case is fine
513 if (btrfs_fs_incompat(fs_info, METADATA_UUID))
514 metadata_uuid = fs_devices->metadata_uuid;
516 metadata_uuid = fs_devices->fsid;
518 if (!memcmp(fsid, metadata_uuid, BTRFS_FSID_SIZE))
521 list_for_each_entry(seed_devs, &fs_devices->seed_list, seed_list)
522 if (!memcmp(fsid, seed_devs->fsid, BTRFS_FSID_SIZE))
528 int btrfs_validate_metadata_buffer(struct btrfs_io_bio *io_bio, u64 phy_offset,
529 struct page *page, u64 start, u64 end,
534 struct extent_buffer *eb;
535 struct btrfs_fs_info *fs_info;
538 u8 result[BTRFS_CSUM_SIZE];
544 eb = (struct extent_buffer *)page->private;
545 fs_info = eb->fs_info;
546 csum_size = fs_info->csum_size;
548 /* the pending IO might have been the only thing that kept this buffer
549 * in memory. Make sure we have a ref for all this other checks
551 atomic_inc(&eb->refs);
553 reads_done = atomic_dec_and_test(&eb->io_pages);
557 eb->read_mirror = mirror;
558 if (test_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags)) {
563 found_start = btrfs_header_bytenr(eb);
564 if (found_start != eb->start) {
565 btrfs_err_rl(fs_info, "bad tree block start, want %llu have %llu",
566 eb->start, found_start);
570 if (check_tree_block_fsid(eb)) {
571 btrfs_err_rl(fs_info, "bad fsid on block %llu",
576 found_level = btrfs_header_level(eb);
577 if (found_level >= BTRFS_MAX_LEVEL) {
578 btrfs_err(fs_info, "bad tree block level %d on %llu",
579 (int)btrfs_header_level(eb), eb->start);
584 btrfs_set_buffer_lockdep_class(btrfs_header_owner(eb),
587 csum_tree_block(eb, result);
589 if (memcmp_extent_buffer(eb, result, 0, csum_size)) {
590 u8 val[BTRFS_CSUM_SIZE] = { 0 };
592 read_extent_buffer(eb, &val, 0, csum_size);
593 btrfs_warn_rl(fs_info,
594 "%s checksum verify failed on %llu wanted " CSUM_FMT " found " CSUM_FMT " level %d",
595 fs_info->sb->s_id, eb->start,
596 CSUM_FMT_VALUE(csum_size, val),
597 CSUM_FMT_VALUE(csum_size, result),
598 btrfs_header_level(eb));
604 * If this is a leaf block and it is corrupt, set the corrupt bit so
605 * that we don't try and read the other copies of this block, just
608 if (found_level == 0 && btrfs_check_leaf_full(eb)) {
609 set_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
613 if (found_level > 0 && btrfs_check_node(eb))
617 set_extent_buffer_uptodate(eb);
620 "block=%llu read time tree block corruption detected",
624 test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags))
625 btree_readahead_hook(eb, ret);
629 * our io error hook is going to dec the io pages
630 * again, we have to make sure it has something
633 atomic_inc(&eb->io_pages);
634 clear_extent_buffer_uptodate(eb);
636 free_extent_buffer(eb);
641 static void end_workqueue_bio(struct bio *bio)
643 struct btrfs_end_io_wq *end_io_wq = bio->bi_private;
644 struct btrfs_fs_info *fs_info;
645 struct btrfs_workqueue *wq;
647 fs_info = end_io_wq->info;
648 end_io_wq->status = bio->bi_status;
650 if (bio_op(bio) == REQ_OP_WRITE) {
651 if (end_io_wq->metadata == BTRFS_WQ_ENDIO_METADATA)
652 wq = fs_info->endio_meta_write_workers;
653 else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_FREE_SPACE)
654 wq = fs_info->endio_freespace_worker;
655 else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56)
656 wq = fs_info->endio_raid56_workers;
658 wq = fs_info->endio_write_workers;
660 if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56)
661 wq = fs_info->endio_raid56_workers;
662 else if (end_io_wq->metadata)
663 wq = fs_info->endio_meta_workers;
665 wq = fs_info->endio_workers;
668 btrfs_init_work(&end_io_wq->work, end_workqueue_fn, NULL, NULL);
669 btrfs_queue_work(wq, &end_io_wq->work);
672 blk_status_t btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio,
673 enum btrfs_wq_endio_type metadata)
675 struct btrfs_end_io_wq *end_io_wq;
677 end_io_wq = kmem_cache_alloc(btrfs_end_io_wq_cache, GFP_NOFS);
679 return BLK_STS_RESOURCE;
681 end_io_wq->private = bio->bi_private;
682 end_io_wq->end_io = bio->bi_end_io;
683 end_io_wq->info = info;
684 end_io_wq->status = 0;
685 end_io_wq->bio = bio;
686 end_io_wq->metadata = metadata;
688 bio->bi_private = end_io_wq;
689 bio->bi_end_io = end_workqueue_bio;
693 static void run_one_async_start(struct btrfs_work *work)
695 struct async_submit_bio *async;
698 async = container_of(work, struct async_submit_bio, work);
699 ret = async->submit_bio_start(async->inode, async->bio, async->bio_offset);
705 * In order to insert checksums into the metadata in large chunks, we wait
706 * until bio submission time. All the pages in the bio are checksummed and
707 * sums are attached onto the ordered extent record.
709 * At IO completion time the csums attached on the ordered extent record are
710 * inserted into the tree.
712 static void run_one_async_done(struct btrfs_work *work)
714 struct async_submit_bio *async;
718 async = container_of(work, struct async_submit_bio, work);
719 inode = async->inode;
721 /* If an error occurred we just want to clean up the bio and move on */
723 async->bio->bi_status = async->status;
724 bio_endio(async->bio);
729 * All of the bios that pass through here are from async helpers.
730 * Use REQ_CGROUP_PUNT to issue them from the owning cgroup's context.
731 * This changes nothing when cgroups aren't in use.
733 async->bio->bi_opf |= REQ_CGROUP_PUNT;
734 ret = btrfs_map_bio(btrfs_sb(inode->i_sb), async->bio, async->mirror_num);
736 async->bio->bi_status = ret;
737 bio_endio(async->bio);
741 static void run_one_async_free(struct btrfs_work *work)
743 struct async_submit_bio *async;
745 async = container_of(work, struct async_submit_bio, work);
749 blk_status_t btrfs_wq_submit_bio(struct inode *inode, struct bio *bio,
750 int mirror_num, unsigned long bio_flags,
752 extent_submit_bio_start_t *submit_bio_start)
754 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
755 struct async_submit_bio *async;
757 async = kmalloc(sizeof(*async), GFP_NOFS);
759 return BLK_STS_RESOURCE;
761 async->inode = inode;
763 async->mirror_num = mirror_num;
764 async->submit_bio_start = submit_bio_start;
766 btrfs_init_work(&async->work, run_one_async_start, run_one_async_done,
769 async->bio_offset = bio_offset;
773 if (op_is_sync(bio->bi_opf))
774 btrfs_set_work_high_priority(&async->work);
776 btrfs_queue_work(fs_info->workers, &async->work);
780 static blk_status_t btree_csum_one_bio(struct bio *bio)
782 struct bio_vec *bvec;
783 struct btrfs_root *root;
785 struct bvec_iter_all iter_all;
787 ASSERT(!bio_flagged(bio, BIO_CLONED));
788 bio_for_each_segment_all(bvec, bio, iter_all) {
789 root = BTRFS_I(bvec->bv_page->mapping->host)->root;
790 ret = csum_dirty_buffer(root->fs_info, bvec->bv_page);
795 return errno_to_blk_status(ret);
798 static blk_status_t btree_submit_bio_start(struct inode *inode, struct bio *bio,
802 * when we're called for a write, we're already in the async
803 * submission context. Just jump into btrfs_map_bio
805 return btree_csum_one_bio(bio);
808 static int check_async_write(struct btrfs_fs_info *fs_info,
809 struct btrfs_inode *bi)
811 if (atomic_read(&bi->sync_writers))
813 if (test_bit(BTRFS_FS_CSUM_IMPL_FAST, &fs_info->flags))
818 blk_status_t btrfs_submit_metadata_bio(struct inode *inode, struct bio *bio,
819 int mirror_num, unsigned long bio_flags)
821 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
822 int async = check_async_write(fs_info, BTRFS_I(inode));
825 if (bio_op(bio) != REQ_OP_WRITE) {
827 * called for a read, do the setup so that checksum validation
828 * can happen in the async kernel threads
830 ret = btrfs_bio_wq_end_io(fs_info, bio,
831 BTRFS_WQ_ENDIO_METADATA);
834 ret = btrfs_map_bio(fs_info, bio, mirror_num);
836 ret = btree_csum_one_bio(bio);
839 ret = btrfs_map_bio(fs_info, bio, mirror_num);
842 * kthread helpers are used to submit writes so that
843 * checksumming can happen in parallel across all CPUs
845 ret = btrfs_wq_submit_bio(inode, bio, mirror_num, 0,
846 0, btree_submit_bio_start);
854 bio->bi_status = ret;
859 #ifdef CONFIG_MIGRATION
860 static int btree_migratepage(struct address_space *mapping,
861 struct page *newpage, struct page *page,
862 enum migrate_mode mode)
865 * we can't safely write a btree page from here,
866 * we haven't done the locking hook
871 * Buffers may be managed in a filesystem specific way.
872 * We must have no buffers or drop them.
874 if (page_has_private(page) &&
875 !try_to_release_page(page, GFP_KERNEL))
877 return migrate_page(mapping, newpage, page, mode);
882 static int btree_writepages(struct address_space *mapping,
883 struct writeback_control *wbc)
885 struct btrfs_fs_info *fs_info;
888 if (wbc->sync_mode == WB_SYNC_NONE) {
890 if (wbc->for_kupdate)
893 fs_info = BTRFS_I(mapping->host)->root->fs_info;
894 /* this is a bit racy, but that's ok */
895 ret = __percpu_counter_compare(&fs_info->dirty_metadata_bytes,
896 BTRFS_DIRTY_METADATA_THRESH,
897 fs_info->dirty_metadata_batch);
901 return btree_write_cache_pages(mapping, wbc);
904 static int btree_releasepage(struct page *page, gfp_t gfp_flags)
906 if (PageWriteback(page) || PageDirty(page))
909 return try_release_extent_buffer(page);
912 static void btree_invalidatepage(struct page *page, unsigned int offset,
915 struct extent_io_tree *tree;
916 tree = &BTRFS_I(page->mapping->host)->io_tree;
917 extent_invalidatepage(tree, page, offset);
918 btree_releasepage(page, GFP_NOFS);
919 if (PagePrivate(page)) {
920 btrfs_warn(BTRFS_I(page->mapping->host)->root->fs_info,
921 "page private not zero on page %llu",
922 (unsigned long long)page_offset(page));
923 detach_page_private(page);
927 static int btree_set_page_dirty(struct page *page)
930 struct extent_buffer *eb;
932 BUG_ON(!PagePrivate(page));
933 eb = (struct extent_buffer *)page->private;
935 BUG_ON(!test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
936 BUG_ON(!atomic_read(&eb->refs));
937 btrfs_assert_tree_locked(eb);
939 return __set_page_dirty_nobuffers(page);
942 static const struct address_space_operations btree_aops = {
943 .writepages = btree_writepages,
944 .releasepage = btree_releasepage,
945 .invalidatepage = btree_invalidatepage,
946 #ifdef CONFIG_MIGRATION
947 .migratepage = btree_migratepage,
949 .set_page_dirty = btree_set_page_dirty,
952 void readahead_tree_block(struct btrfs_fs_info *fs_info, u64 bytenr)
954 struct extent_buffer *buf = NULL;
957 buf = btrfs_find_create_tree_block(fs_info, bytenr);
961 ret = read_extent_buffer_pages(buf, WAIT_NONE, 0);
963 free_extent_buffer_stale(buf);
965 free_extent_buffer(buf);
968 struct extent_buffer *btrfs_find_create_tree_block(
969 struct btrfs_fs_info *fs_info,
972 if (btrfs_is_testing(fs_info))
973 return alloc_test_extent_buffer(fs_info, bytenr);
974 return alloc_extent_buffer(fs_info, bytenr);
978 * Read tree block at logical address @bytenr and do variant basic but critical
981 * @parent_transid: expected transid of this tree block, skip check if 0
982 * @level: expected level, mandatory check
983 * @first_key: expected key in slot 0, skip check if NULL
985 struct extent_buffer *read_tree_block(struct btrfs_fs_info *fs_info, u64 bytenr,
986 u64 parent_transid, int level,
987 struct btrfs_key *first_key)
989 struct extent_buffer *buf = NULL;
992 buf = btrfs_find_create_tree_block(fs_info, bytenr);
996 ret = btree_read_extent_buffer_pages(buf, parent_transid,
999 free_extent_buffer_stale(buf);
1000 return ERR_PTR(ret);
1006 void btrfs_clean_tree_block(struct extent_buffer *buf)
1008 struct btrfs_fs_info *fs_info = buf->fs_info;
1009 if (btrfs_header_generation(buf) ==
1010 fs_info->running_transaction->transid) {
1011 btrfs_assert_tree_locked(buf);
1013 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)) {
1014 percpu_counter_add_batch(&fs_info->dirty_metadata_bytes,
1016 fs_info->dirty_metadata_batch);
1017 /* ugh, clear_extent_buffer_dirty needs to lock the page */
1018 btrfs_set_lock_blocking_write(buf);
1019 clear_extent_buffer_dirty(buf);
1024 static void __setup_root(struct btrfs_root *root, struct btrfs_fs_info *fs_info,
1027 bool dummy = test_bit(BTRFS_FS_STATE_DUMMY_FS_INFO, &fs_info->fs_state);
1028 root->fs_info = fs_info;
1030 root->commit_root = NULL;
1032 root->orphan_cleanup_state = 0;
1034 root->last_trans = 0;
1035 root->highest_objectid = 0;
1036 root->nr_delalloc_inodes = 0;
1037 root->nr_ordered_extents = 0;
1038 root->inode_tree = RB_ROOT;
1039 INIT_RADIX_TREE(&root->delayed_nodes_tree, GFP_ATOMIC);
1040 root->block_rsv = NULL;
1042 INIT_LIST_HEAD(&root->dirty_list);
1043 INIT_LIST_HEAD(&root->root_list);
1044 INIT_LIST_HEAD(&root->delalloc_inodes);
1045 INIT_LIST_HEAD(&root->delalloc_root);
1046 INIT_LIST_HEAD(&root->ordered_extents);
1047 INIT_LIST_HEAD(&root->ordered_root);
1048 INIT_LIST_HEAD(&root->reloc_dirty_list);
1049 INIT_LIST_HEAD(&root->logged_list[0]);
1050 INIT_LIST_HEAD(&root->logged_list[1]);
1051 spin_lock_init(&root->inode_lock);
1052 spin_lock_init(&root->delalloc_lock);
1053 spin_lock_init(&root->ordered_extent_lock);
1054 spin_lock_init(&root->accounting_lock);
1055 spin_lock_init(&root->log_extents_lock[0]);
1056 spin_lock_init(&root->log_extents_lock[1]);
1057 spin_lock_init(&root->qgroup_meta_rsv_lock);
1058 mutex_init(&root->objectid_mutex);
1059 mutex_init(&root->log_mutex);
1060 mutex_init(&root->ordered_extent_mutex);
1061 mutex_init(&root->delalloc_mutex);
1062 init_waitqueue_head(&root->qgroup_flush_wait);
1063 init_waitqueue_head(&root->log_writer_wait);
1064 init_waitqueue_head(&root->log_commit_wait[0]);
1065 init_waitqueue_head(&root->log_commit_wait[1]);
1066 INIT_LIST_HEAD(&root->log_ctxs[0]);
1067 INIT_LIST_HEAD(&root->log_ctxs[1]);
1068 atomic_set(&root->log_commit[0], 0);
1069 atomic_set(&root->log_commit[1], 0);
1070 atomic_set(&root->log_writers, 0);
1071 atomic_set(&root->log_batch, 0);
1072 refcount_set(&root->refs, 1);
1073 atomic_set(&root->snapshot_force_cow, 0);
1074 atomic_set(&root->nr_swapfiles, 0);
1075 root->log_transid = 0;
1076 root->log_transid_committed = -1;
1077 root->last_log_commit = 0;
1079 extent_io_tree_init(fs_info, &root->dirty_log_pages,
1080 IO_TREE_ROOT_DIRTY_LOG_PAGES, NULL);
1081 extent_io_tree_init(fs_info, &root->log_csum_range,
1082 IO_TREE_LOG_CSUM_RANGE, NULL);
1085 memset(&root->root_key, 0, sizeof(root->root_key));
1086 memset(&root->root_item, 0, sizeof(root->root_item));
1087 memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
1088 root->root_key.objectid = objectid;
1091 spin_lock_init(&root->root_item_lock);
1092 btrfs_qgroup_init_swapped_blocks(&root->swapped_blocks);
1093 #ifdef CONFIG_BTRFS_DEBUG
1094 INIT_LIST_HEAD(&root->leak_list);
1095 spin_lock(&fs_info->fs_roots_radix_lock);
1096 list_add_tail(&root->leak_list, &fs_info->allocated_roots);
1097 spin_unlock(&fs_info->fs_roots_radix_lock);
1101 static struct btrfs_root *btrfs_alloc_root(struct btrfs_fs_info *fs_info,
1102 u64 objectid, gfp_t flags)
1104 struct btrfs_root *root = kzalloc(sizeof(*root), flags);
1106 __setup_root(root, fs_info, objectid);
1110 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
1111 /* Should only be used by the testing infrastructure */
1112 struct btrfs_root *btrfs_alloc_dummy_root(struct btrfs_fs_info *fs_info)
1114 struct btrfs_root *root;
1117 return ERR_PTR(-EINVAL);
1119 root = btrfs_alloc_root(fs_info, BTRFS_ROOT_TREE_OBJECTID, GFP_KERNEL);
1121 return ERR_PTR(-ENOMEM);
1123 /* We don't use the stripesize in selftest, set it as sectorsize */
1124 root->alloc_bytenr = 0;
1130 struct btrfs_root *btrfs_create_tree(struct btrfs_trans_handle *trans,
1133 struct btrfs_fs_info *fs_info = trans->fs_info;
1134 struct extent_buffer *leaf;
1135 struct btrfs_root *tree_root = fs_info->tree_root;
1136 struct btrfs_root *root;
1137 struct btrfs_key key;
1138 unsigned int nofs_flag;
1142 * We're holding a transaction handle, so use a NOFS memory allocation
1143 * context to avoid deadlock if reclaim happens.
1145 nofs_flag = memalloc_nofs_save();
1146 root = btrfs_alloc_root(fs_info, objectid, GFP_KERNEL);
1147 memalloc_nofs_restore(nofs_flag);
1149 return ERR_PTR(-ENOMEM);
1151 root->root_key.objectid = objectid;
1152 root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1153 root->root_key.offset = 0;
1155 leaf = btrfs_alloc_tree_block(trans, root, 0, objectid, NULL, 0, 0, 0,
1156 BTRFS_NESTING_NORMAL);
1158 ret = PTR_ERR(leaf);
1164 btrfs_mark_buffer_dirty(leaf);
1166 root->commit_root = btrfs_root_node(root);
1167 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
1169 btrfs_set_root_flags(&root->root_item, 0);
1170 btrfs_set_root_limit(&root->root_item, 0);
1171 btrfs_set_root_bytenr(&root->root_item, leaf->start);
1172 btrfs_set_root_generation(&root->root_item, trans->transid);
1173 btrfs_set_root_level(&root->root_item, 0);
1174 btrfs_set_root_refs(&root->root_item, 1);
1175 btrfs_set_root_used(&root->root_item, leaf->len);
1176 btrfs_set_root_last_snapshot(&root->root_item, 0);
1177 btrfs_set_root_dirid(&root->root_item, 0);
1178 if (is_fstree(objectid))
1179 generate_random_guid(root->root_item.uuid);
1181 export_guid(root->root_item.uuid, &guid_null);
1182 btrfs_set_root_drop_level(&root->root_item, 0);
1184 key.objectid = objectid;
1185 key.type = BTRFS_ROOT_ITEM_KEY;
1187 ret = btrfs_insert_root(trans, tree_root, &key, &root->root_item);
1191 btrfs_tree_unlock(leaf);
1197 btrfs_tree_unlock(leaf);
1198 btrfs_put_root(root);
1200 return ERR_PTR(ret);
1203 static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans,
1204 struct btrfs_fs_info *fs_info)
1206 struct btrfs_root *root;
1207 struct extent_buffer *leaf;
1209 root = btrfs_alloc_root(fs_info, BTRFS_TREE_LOG_OBJECTID, GFP_NOFS);
1211 return ERR_PTR(-ENOMEM);
1213 root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
1214 root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1215 root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
1218 * DON'T set SHAREABLE bit for log trees.
1220 * Log trees are not exposed to user space thus can't be snapshotted,
1221 * and they go away before a real commit is actually done.
1223 * They do store pointers to file data extents, and those reference
1224 * counts still get updated (along with back refs to the log tree).
1227 leaf = btrfs_alloc_tree_block(trans, root, 0, BTRFS_TREE_LOG_OBJECTID,
1228 NULL, 0, 0, 0, BTRFS_NESTING_NORMAL);
1230 btrfs_put_root(root);
1231 return ERR_CAST(leaf);
1236 btrfs_mark_buffer_dirty(root->node);
1237 btrfs_tree_unlock(root->node);
1241 int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
1242 struct btrfs_fs_info *fs_info)
1244 struct btrfs_root *log_root;
1246 log_root = alloc_log_tree(trans, fs_info);
1247 if (IS_ERR(log_root))
1248 return PTR_ERR(log_root);
1249 WARN_ON(fs_info->log_root_tree);
1250 fs_info->log_root_tree = log_root;
1254 int btrfs_add_log_tree(struct btrfs_trans_handle *trans,
1255 struct btrfs_root *root)
1257 struct btrfs_fs_info *fs_info = root->fs_info;
1258 struct btrfs_root *log_root;
1259 struct btrfs_inode_item *inode_item;
1261 log_root = alloc_log_tree(trans, fs_info);
1262 if (IS_ERR(log_root))
1263 return PTR_ERR(log_root);
1265 log_root->last_trans = trans->transid;
1266 log_root->root_key.offset = root->root_key.objectid;
1268 inode_item = &log_root->root_item.inode;
1269 btrfs_set_stack_inode_generation(inode_item, 1);
1270 btrfs_set_stack_inode_size(inode_item, 3);
1271 btrfs_set_stack_inode_nlink(inode_item, 1);
1272 btrfs_set_stack_inode_nbytes(inode_item,
1274 btrfs_set_stack_inode_mode(inode_item, S_IFDIR | 0755);
1276 btrfs_set_root_node(&log_root->root_item, log_root->node);
1278 WARN_ON(root->log_root);
1279 root->log_root = log_root;
1280 root->log_transid = 0;
1281 root->log_transid_committed = -1;
1282 root->last_log_commit = 0;
1286 static struct btrfs_root *read_tree_root_path(struct btrfs_root *tree_root,
1287 struct btrfs_path *path,
1288 struct btrfs_key *key)
1290 struct btrfs_root *root;
1291 struct btrfs_fs_info *fs_info = tree_root->fs_info;
1296 root = btrfs_alloc_root(fs_info, key->objectid, GFP_NOFS);
1298 return ERR_PTR(-ENOMEM);
1300 ret = btrfs_find_root(tree_root, key, path,
1301 &root->root_item, &root->root_key);
1308 generation = btrfs_root_generation(&root->root_item);
1309 level = btrfs_root_level(&root->root_item);
1310 root->node = read_tree_block(fs_info,
1311 btrfs_root_bytenr(&root->root_item),
1312 generation, level, NULL);
1313 if (IS_ERR(root->node)) {
1314 ret = PTR_ERR(root->node);
1317 } else if (!btrfs_buffer_uptodate(root->node, generation, 0)) {
1321 root->commit_root = btrfs_root_node(root);
1324 btrfs_put_root(root);
1325 return ERR_PTR(ret);
1328 struct btrfs_root *btrfs_read_tree_root(struct btrfs_root *tree_root,
1329 struct btrfs_key *key)
1331 struct btrfs_root *root;
1332 struct btrfs_path *path;
1334 path = btrfs_alloc_path();
1336 return ERR_PTR(-ENOMEM);
1337 root = read_tree_root_path(tree_root, path, key);
1338 btrfs_free_path(path);
1344 * Initialize subvolume root in-memory structure
1346 * @anon_dev: anonymous device to attach to the root, if zero, allocate new
1348 static int btrfs_init_fs_root(struct btrfs_root *root, dev_t anon_dev)
1351 unsigned int nofs_flag;
1353 root->free_ino_ctl = kzalloc(sizeof(*root->free_ino_ctl), GFP_NOFS);
1354 root->free_ino_pinned = kzalloc(sizeof(*root->free_ino_pinned),
1356 if (!root->free_ino_pinned || !root->free_ino_ctl) {
1362 * We might be called under a transaction (e.g. indirect backref
1363 * resolution) which could deadlock if it triggers memory reclaim
1365 nofs_flag = memalloc_nofs_save();
1366 ret = btrfs_drew_lock_init(&root->snapshot_lock);
1367 memalloc_nofs_restore(nofs_flag);
1371 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID &&
1372 root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID) {
1373 set_bit(BTRFS_ROOT_SHAREABLE, &root->state);
1374 btrfs_check_and_init_root_item(&root->root_item);
1377 btrfs_init_free_ino_ctl(root);
1378 spin_lock_init(&root->ino_cache_lock);
1379 init_waitqueue_head(&root->ino_cache_wait);
1382 * Don't assign anonymous block device to roots that are not exposed to
1383 * userspace, the id pool is limited to 1M
1385 if (is_fstree(root->root_key.objectid) &&
1386 btrfs_root_refs(&root->root_item) > 0) {
1388 ret = get_anon_bdev(&root->anon_dev);
1392 root->anon_dev = anon_dev;
1396 mutex_lock(&root->objectid_mutex);
1397 ret = btrfs_find_highest_objectid(root,
1398 &root->highest_objectid);
1400 mutex_unlock(&root->objectid_mutex);
1404 ASSERT(root->highest_objectid <= BTRFS_LAST_FREE_OBJECTID);
1406 mutex_unlock(&root->objectid_mutex);
1410 /* The caller is responsible to call btrfs_free_fs_root */
1414 static struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info,
1417 struct btrfs_root *root;
1419 spin_lock(&fs_info->fs_roots_radix_lock);
1420 root = radix_tree_lookup(&fs_info->fs_roots_radix,
1421 (unsigned long)root_id);
1423 root = btrfs_grab_root(root);
1424 spin_unlock(&fs_info->fs_roots_radix_lock);
1428 static struct btrfs_root *btrfs_get_global_root(struct btrfs_fs_info *fs_info,
1431 if (objectid == BTRFS_ROOT_TREE_OBJECTID)
1432 return btrfs_grab_root(fs_info->tree_root);
1433 if (objectid == BTRFS_EXTENT_TREE_OBJECTID)
1434 return btrfs_grab_root(fs_info->extent_root);
1435 if (objectid == BTRFS_CHUNK_TREE_OBJECTID)
1436 return btrfs_grab_root(fs_info->chunk_root);
1437 if (objectid == BTRFS_DEV_TREE_OBJECTID)
1438 return btrfs_grab_root(fs_info->dev_root);
1439 if (objectid == BTRFS_CSUM_TREE_OBJECTID)
1440 return btrfs_grab_root(fs_info->csum_root);
1441 if (objectid == BTRFS_QUOTA_TREE_OBJECTID)
1442 return btrfs_grab_root(fs_info->quota_root) ?
1443 fs_info->quota_root : ERR_PTR(-ENOENT);
1444 if (objectid == BTRFS_UUID_TREE_OBJECTID)
1445 return btrfs_grab_root(fs_info->uuid_root) ?
1446 fs_info->uuid_root : ERR_PTR(-ENOENT);
1447 if (objectid == BTRFS_FREE_SPACE_TREE_OBJECTID)
1448 return btrfs_grab_root(fs_info->free_space_root) ?
1449 fs_info->free_space_root : ERR_PTR(-ENOENT);
1453 int btrfs_insert_fs_root(struct btrfs_fs_info *fs_info,
1454 struct btrfs_root *root)
1458 ret = radix_tree_preload(GFP_NOFS);
1462 spin_lock(&fs_info->fs_roots_radix_lock);
1463 ret = radix_tree_insert(&fs_info->fs_roots_radix,
1464 (unsigned long)root->root_key.objectid,
1467 btrfs_grab_root(root);
1468 set_bit(BTRFS_ROOT_IN_RADIX, &root->state);
1470 spin_unlock(&fs_info->fs_roots_radix_lock);
1471 radix_tree_preload_end();
1476 void btrfs_check_leaked_roots(struct btrfs_fs_info *fs_info)
1478 #ifdef CONFIG_BTRFS_DEBUG
1479 struct btrfs_root *root;
1481 while (!list_empty(&fs_info->allocated_roots)) {
1482 char buf[BTRFS_ROOT_NAME_BUF_LEN];
1484 root = list_first_entry(&fs_info->allocated_roots,
1485 struct btrfs_root, leak_list);
1486 btrfs_err(fs_info, "leaked root %s refcount %d",
1487 btrfs_root_name(root->root_key.objectid, buf),
1488 refcount_read(&root->refs));
1489 while (refcount_read(&root->refs) > 1)
1490 btrfs_put_root(root);
1491 btrfs_put_root(root);
1496 void btrfs_free_fs_info(struct btrfs_fs_info *fs_info)
1498 percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
1499 percpu_counter_destroy(&fs_info->delalloc_bytes);
1500 percpu_counter_destroy(&fs_info->dio_bytes);
1501 percpu_counter_destroy(&fs_info->dev_replace.bio_counter);
1502 btrfs_free_csum_hash(fs_info);
1503 btrfs_free_stripe_hash_table(fs_info);
1504 btrfs_free_ref_cache(fs_info);
1505 kfree(fs_info->balance_ctl);
1506 kfree(fs_info->delayed_root);
1507 btrfs_put_root(fs_info->extent_root);
1508 btrfs_put_root(fs_info->tree_root);
1509 btrfs_put_root(fs_info->chunk_root);
1510 btrfs_put_root(fs_info->dev_root);
1511 btrfs_put_root(fs_info->csum_root);
1512 btrfs_put_root(fs_info->quota_root);
1513 btrfs_put_root(fs_info->uuid_root);
1514 btrfs_put_root(fs_info->free_space_root);
1515 btrfs_put_root(fs_info->fs_root);
1516 btrfs_put_root(fs_info->data_reloc_root);
1517 btrfs_check_leaked_roots(fs_info);
1518 btrfs_extent_buffer_leak_debug_check(fs_info);
1519 kfree(fs_info->super_copy);
1520 kfree(fs_info->super_for_commit);
1526 * Get an in-memory reference of a root structure.
1528 * For essential trees like root/extent tree, we grab it from fs_info directly.
1529 * For subvolume trees, we check the cached filesystem roots first. If not
1530 * found, then read it from disk and add it to cached fs roots.
1532 * Caller should release the root by calling btrfs_put_root() after the usage.
1534 * NOTE: Reloc and log trees can't be read by this function as they share the
1535 * same root objectid.
1537 * @objectid: root id
1538 * @anon_dev: preallocated anonymous block device number for new roots,
1539 * pass 0 for new allocation.
1540 * @check_ref: whether to check root item references, If true, return -ENOENT
1543 static struct btrfs_root *btrfs_get_root_ref(struct btrfs_fs_info *fs_info,
1544 u64 objectid, dev_t anon_dev,
1547 struct btrfs_root *root;
1548 struct btrfs_path *path;
1549 struct btrfs_key key;
1552 root = btrfs_get_global_root(fs_info, objectid);
1556 root = btrfs_lookup_fs_root(fs_info, objectid);
1558 /* Shouldn't get preallocated anon_dev for cached roots */
1560 if (check_ref && btrfs_root_refs(&root->root_item) == 0) {
1561 btrfs_put_root(root);
1562 return ERR_PTR(-ENOENT);
1567 key.objectid = objectid;
1568 key.type = BTRFS_ROOT_ITEM_KEY;
1569 key.offset = (u64)-1;
1570 root = btrfs_read_tree_root(fs_info->tree_root, &key);
1574 if (check_ref && btrfs_root_refs(&root->root_item) == 0) {
1579 ret = btrfs_init_fs_root(root, anon_dev);
1583 path = btrfs_alloc_path();
1588 key.objectid = BTRFS_ORPHAN_OBJECTID;
1589 key.type = BTRFS_ORPHAN_ITEM_KEY;
1590 key.offset = objectid;
1592 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
1593 btrfs_free_path(path);
1597 set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state);
1599 ret = btrfs_insert_fs_root(fs_info, root);
1601 btrfs_put_root(root);
1608 btrfs_put_root(root);
1609 return ERR_PTR(ret);
1613 * Get in-memory reference of a root structure
1615 * @objectid: tree objectid
1616 * @check_ref: if set, verify that the tree exists and the item has at least
1619 struct btrfs_root *btrfs_get_fs_root(struct btrfs_fs_info *fs_info,
1620 u64 objectid, bool check_ref)
1622 return btrfs_get_root_ref(fs_info, objectid, 0, check_ref);
1626 * Get in-memory reference of a root structure, created as new, optionally pass
1627 * the anonymous block device id
1629 * @objectid: tree objectid
1630 * @anon_dev: if zero, allocate a new anonymous block device or use the
1633 struct btrfs_root *btrfs_get_new_fs_root(struct btrfs_fs_info *fs_info,
1634 u64 objectid, dev_t anon_dev)
1636 return btrfs_get_root_ref(fs_info, objectid, anon_dev, true);
1640 * btrfs_get_fs_root_commit_root - return a root for the given objectid
1641 * @fs_info: the fs_info
1642 * @objectid: the objectid we need to lookup
1644 * This is exclusively used for backref walking, and exists specifically because
1645 * of how qgroups does lookups. Qgroups will do a backref lookup at delayed ref
1646 * creation time, which means we may have to read the tree_root in order to look
1647 * up a fs root that is not in memory. If the root is not in memory we will
1648 * read the tree root commit root and look up the fs root from there. This is a
1649 * temporary root, it will not be inserted into the radix tree as it doesn't
1650 * have the most uptodate information, it'll simply be discarded once the
1651 * backref code is finished using the root.
1653 struct btrfs_root *btrfs_get_fs_root_commit_root(struct btrfs_fs_info *fs_info,
1654 struct btrfs_path *path,
1657 struct btrfs_root *root;
1658 struct btrfs_key key;
1660 ASSERT(path->search_commit_root && path->skip_locking);
1663 * This can return -ENOENT if we ask for a root that doesn't exist, but
1664 * since this is called via the backref walking code we won't be looking
1665 * up a root that doesn't exist, unless there's corruption. So if root
1666 * != NULL just return it.
1668 root = btrfs_get_global_root(fs_info, objectid);
1672 root = btrfs_lookup_fs_root(fs_info, objectid);
1676 key.objectid = objectid;
1677 key.type = BTRFS_ROOT_ITEM_KEY;
1678 key.offset = (u64)-1;
1679 root = read_tree_root_path(fs_info->tree_root, path, &key);
1680 btrfs_release_path(path);
1686 * called by the kthread helper functions to finally call the bio end_io
1687 * functions. This is where read checksum verification actually happens
1689 static void end_workqueue_fn(struct btrfs_work *work)
1692 struct btrfs_end_io_wq *end_io_wq;
1694 end_io_wq = container_of(work, struct btrfs_end_io_wq, work);
1695 bio = end_io_wq->bio;
1697 bio->bi_status = end_io_wq->status;
1698 bio->bi_private = end_io_wq->private;
1699 bio->bi_end_io = end_io_wq->end_io;
1701 kmem_cache_free(btrfs_end_io_wq_cache, end_io_wq);
1704 static int cleaner_kthread(void *arg)
1706 struct btrfs_root *root = arg;
1707 struct btrfs_fs_info *fs_info = root->fs_info;
1713 set_bit(BTRFS_FS_CLEANER_RUNNING, &fs_info->flags);
1715 /* Make the cleaner go to sleep early. */
1716 if (btrfs_need_cleaner_sleep(fs_info))
1720 * Do not do anything if we might cause open_ctree() to block
1721 * before we have finished mounting the filesystem.
1723 if (!test_bit(BTRFS_FS_OPEN, &fs_info->flags))
1726 if (!mutex_trylock(&fs_info->cleaner_mutex))
1730 * Avoid the problem that we change the status of the fs
1731 * during the above check and trylock.
1733 if (btrfs_need_cleaner_sleep(fs_info)) {
1734 mutex_unlock(&fs_info->cleaner_mutex);
1738 btrfs_run_delayed_iputs(fs_info);
1740 again = btrfs_clean_one_deleted_snapshot(root);
1741 mutex_unlock(&fs_info->cleaner_mutex);
1744 * The defragger has dealt with the R/O remount and umount,
1745 * needn't do anything special here.
1747 btrfs_run_defrag_inodes(fs_info);
1750 * Acquires fs_info->delete_unused_bgs_mutex to avoid racing
1751 * with relocation (btrfs_relocate_chunk) and relocation
1752 * acquires fs_info->cleaner_mutex (btrfs_relocate_block_group)
1753 * after acquiring fs_info->delete_unused_bgs_mutex. So we
1754 * can't hold, nor need to, fs_info->cleaner_mutex when deleting
1755 * unused block groups.
1757 btrfs_delete_unused_bgs(fs_info);
1759 clear_bit(BTRFS_FS_CLEANER_RUNNING, &fs_info->flags);
1760 if (kthread_should_park())
1762 if (kthread_should_stop())
1765 set_current_state(TASK_INTERRUPTIBLE);
1767 __set_current_state(TASK_RUNNING);
1772 static int transaction_kthread(void *arg)
1774 struct btrfs_root *root = arg;
1775 struct btrfs_fs_info *fs_info = root->fs_info;
1776 struct btrfs_trans_handle *trans;
1777 struct btrfs_transaction *cur;
1780 unsigned long delay;
1784 cannot_commit = false;
1785 delay = msecs_to_jiffies(fs_info->commit_interval * 1000);
1786 mutex_lock(&fs_info->transaction_kthread_mutex);
1788 spin_lock(&fs_info->trans_lock);
1789 cur = fs_info->running_transaction;
1791 spin_unlock(&fs_info->trans_lock);
1795 delta = ktime_get_seconds() - cur->start_time;
1796 if (cur->state < TRANS_STATE_COMMIT_START &&
1797 delta < fs_info->commit_interval) {
1798 spin_unlock(&fs_info->trans_lock);
1799 delay -= msecs_to_jiffies((delta - 1) * 1000);
1801 msecs_to_jiffies(fs_info->commit_interval * 1000));
1804 transid = cur->transid;
1805 spin_unlock(&fs_info->trans_lock);
1807 /* If the file system is aborted, this will always fail. */
1808 trans = btrfs_attach_transaction(root);
1809 if (IS_ERR(trans)) {
1810 if (PTR_ERR(trans) != -ENOENT)
1811 cannot_commit = true;
1814 if (transid == trans->transid) {
1815 btrfs_commit_transaction(trans);
1817 btrfs_end_transaction(trans);
1820 wake_up_process(fs_info->cleaner_kthread);
1821 mutex_unlock(&fs_info->transaction_kthread_mutex);
1823 if (unlikely(test_bit(BTRFS_FS_STATE_ERROR,
1824 &fs_info->fs_state)))
1825 btrfs_cleanup_transaction(fs_info);
1826 if (!kthread_should_stop() &&
1827 (!btrfs_transaction_blocked(fs_info) ||
1829 schedule_timeout_interruptible(delay);
1830 } while (!kthread_should_stop());
1835 * This will find the highest generation in the array of root backups. The
1836 * index of the highest array is returned, or -EINVAL if we can't find
1839 * We check to make sure the array is valid by comparing the
1840 * generation of the latest root in the array with the generation
1841 * in the super block. If they don't match we pitch it.
1843 static int find_newest_super_backup(struct btrfs_fs_info *info)
1845 const u64 newest_gen = btrfs_super_generation(info->super_copy);
1847 struct btrfs_root_backup *root_backup;
1850 for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) {
1851 root_backup = info->super_copy->super_roots + i;
1852 cur = btrfs_backup_tree_root_gen(root_backup);
1853 if (cur == newest_gen)
1861 * copy all the root pointers into the super backup array.
1862 * this will bump the backup pointer by one when it is
1865 static void backup_super_roots(struct btrfs_fs_info *info)
1867 const int next_backup = info->backup_root_index;
1868 struct btrfs_root_backup *root_backup;
1870 root_backup = info->super_for_commit->super_roots + next_backup;
1873 * make sure all of our padding and empty slots get zero filled
1874 * regardless of which ones we use today
1876 memset(root_backup, 0, sizeof(*root_backup));
1878 info->backup_root_index = (next_backup + 1) % BTRFS_NUM_BACKUP_ROOTS;
1880 btrfs_set_backup_tree_root(root_backup, info->tree_root->node->start);
1881 btrfs_set_backup_tree_root_gen(root_backup,
1882 btrfs_header_generation(info->tree_root->node));
1884 btrfs_set_backup_tree_root_level(root_backup,
1885 btrfs_header_level(info->tree_root->node));
1887 btrfs_set_backup_chunk_root(root_backup, info->chunk_root->node->start);
1888 btrfs_set_backup_chunk_root_gen(root_backup,
1889 btrfs_header_generation(info->chunk_root->node));
1890 btrfs_set_backup_chunk_root_level(root_backup,
1891 btrfs_header_level(info->chunk_root->node));
1893 btrfs_set_backup_extent_root(root_backup, info->extent_root->node->start);
1894 btrfs_set_backup_extent_root_gen(root_backup,
1895 btrfs_header_generation(info->extent_root->node));
1896 btrfs_set_backup_extent_root_level(root_backup,
1897 btrfs_header_level(info->extent_root->node));
1900 * we might commit during log recovery, which happens before we set
1901 * the fs_root. Make sure it is valid before we fill it in.
1903 if (info->fs_root && info->fs_root->node) {
1904 btrfs_set_backup_fs_root(root_backup,
1905 info->fs_root->node->start);
1906 btrfs_set_backup_fs_root_gen(root_backup,
1907 btrfs_header_generation(info->fs_root->node));
1908 btrfs_set_backup_fs_root_level(root_backup,
1909 btrfs_header_level(info->fs_root->node));
1912 btrfs_set_backup_dev_root(root_backup, info->dev_root->node->start);
1913 btrfs_set_backup_dev_root_gen(root_backup,
1914 btrfs_header_generation(info->dev_root->node));
1915 btrfs_set_backup_dev_root_level(root_backup,
1916 btrfs_header_level(info->dev_root->node));
1918 btrfs_set_backup_csum_root(root_backup, info->csum_root->node->start);
1919 btrfs_set_backup_csum_root_gen(root_backup,
1920 btrfs_header_generation(info->csum_root->node));
1921 btrfs_set_backup_csum_root_level(root_backup,
1922 btrfs_header_level(info->csum_root->node));
1924 btrfs_set_backup_total_bytes(root_backup,
1925 btrfs_super_total_bytes(info->super_copy));
1926 btrfs_set_backup_bytes_used(root_backup,
1927 btrfs_super_bytes_used(info->super_copy));
1928 btrfs_set_backup_num_devices(root_backup,
1929 btrfs_super_num_devices(info->super_copy));
1932 * if we don't copy this out to the super_copy, it won't get remembered
1933 * for the next commit
1935 memcpy(&info->super_copy->super_roots,
1936 &info->super_for_commit->super_roots,
1937 sizeof(*root_backup) * BTRFS_NUM_BACKUP_ROOTS);
1941 * read_backup_root - Reads a backup root based on the passed priority. Prio 0
1942 * is the newest, prio 1/2/3 are 2nd newest/3rd newest/4th (oldest) backup roots
1944 * fs_info - filesystem whose backup roots need to be read
1945 * priority - priority of backup root required
1947 * Returns backup root index on success and -EINVAL otherwise.
1949 static int read_backup_root(struct btrfs_fs_info *fs_info, u8 priority)
1951 int backup_index = find_newest_super_backup(fs_info);
1952 struct btrfs_super_block *super = fs_info->super_copy;
1953 struct btrfs_root_backup *root_backup;
1955 if (priority < BTRFS_NUM_BACKUP_ROOTS && backup_index >= 0) {
1957 return backup_index;
1959 backup_index = backup_index + BTRFS_NUM_BACKUP_ROOTS - priority;
1960 backup_index %= BTRFS_NUM_BACKUP_ROOTS;
1965 root_backup = super->super_roots + backup_index;
1967 btrfs_set_super_generation(super,
1968 btrfs_backup_tree_root_gen(root_backup));
1969 btrfs_set_super_root(super, btrfs_backup_tree_root(root_backup));
1970 btrfs_set_super_root_level(super,
1971 btrfs_backup_tree_root_level(root_backup));
1972 btrfs_set_super_bytes_used(super, btrfs_backup_bytes_used(root_backup));
1975 * Fixme: the total bytes and num_devices need to match or we should
1978 btrfs_set_super_total_bytes(super, btrfs_backup_total_bytes(root_backup));
1979 btrfs_set_super_num_devices(super, btrfs_backup_num_devices(root_backup));
1981 return backup_index;
1984 /* helper to cleanup workers */
1985 static void btrfs_stop_all_workers(struct btrfs_fs_info *fs_info)
1987 btrfs_destroy_workqueue(fs_info->fixup_workers);
1988 btrfs_destroy_workqueue(fs_info->delalloc_workers);
1989 btrfs_destroy_workqueue(fs_info->workers);
1990 btrfs_destroy_workqueue(fs_info->endio_workers);
1991 btrfs_destroy_workqueue(fs_info->endio_raid56_workers);
1992 btrfs_destroy_workqueue(fs_info->rmw_workers);
1993 btrfs_destroy_workqueue(fs_info->endio_write_workers);
1994 btrfs_destroy_workqueue(fs_info->endio_freespace_worker);
1995 btrfs_destroy_workqueue(fs_info->delayed_workers);
1996 btrfs_destroy_workqueue(fs_info->caching_workers);
1997 btrfs_destroy_workqueue(fs_info->readahead_workers);
1998 btrfs_destroy_workqueue(fs_info->flush_workers);
1999 btrfs_destroy_workqueue(fs_info->qgroup_rescan_workers);
2000 if (fs_info->discard_ctl.discard_workers)
2001 destroy_workqueue(fs_info->discard_ctl.discard_workers);
2003 * Now that all other work queues are destroyed, we can safely destroy
2004 * the queues used for metadata I/O, since tasks from those other work
2005 * queues can do metadata I/O operations.
2007 btrfs_destroy_workqueue(fs_info->endio_meta_workers);
2008 btrfs_destroy_workqueue(fs_info->endio_meta_write_workers);
2011 static void free_root_extent_buffers(struct btrfs_root *root)
2014 free_extent_buffer(root->node);
2015 free_extent_buffer(root->commit_root);
2017 root->commit_root = NULL;
2021 /* helper to cleanup tree roots */
2022 static void free_root_pointers(struct btrfs_fs_info *info, bool free_chunk_root)
2024 free_root_extent_buffers(info->tree_root);
2026 free_root_extent_buffers(info->dev_root);
2027 free_root_extent_buffers(info->extent_root);
2028 free_root_extent_buffers(info->csum_root);
2029 free_root_extent_buffers(info->quota_root);
2030 free_root_extent_buffers(info->uuid_root);
2031 free_root_extent_buffers(info->fs_root);
2032 free_root_extent_buffers(info->data_reloc_root);
2033 if (free_chunk_root)
2034 free_root_extent_buffers(info->chunk_root);
2035 free_root_extent_buffers(info->free_space_root);
2038 void btrfs_put_root(struct btrfs_root *root)
2043 if (refcount_dec_and_test(&root->refs)) {
2044 WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree));
2045 WARN_ON(test_bit(BTRFS_ROOT_DEAD_RELOC_TREE, &root->state));
2047 free_anon_bdev(root->anon_dev);
2048 btrfs_drew_lock_destroy(&root->snapshot_lock);
2049 free_root_extent_buffers(root);
2050 kfree(root->free_ino_ctl);
2051 kfree(root->free_ino_pinned);
2052 #ifdef CONFIG_BTRFS_DEBUG
2053 spin_lock(&root->fs_info->fs_roots_radix_lock);
2054 list_del_init(&root->leak_list);
2055 spin_unlock(&root->fs_info->fs_roots_radix_lock);
2061 void btrfs_free_fs_roots(struct btrfs_fs_info *fs_info)
2064 struct btrfs_root *gang[8];
2067 while (!list_empty(&fs_info->dead_roots)) {
2068 gang[0] = list_entry(fs_info->dead_roots.next,
2069 struct btrfs_root, root_list);
2070 list_del(&gang[0]->root_list);
2072 if (test_bit(BTRFS_ROOT_IN_RADIX, &gang[0]->state))
2073 btrfs_drop_and_free_fs_root(fs_info, gang[0]);
2074 btrfs_put_root(gang[0]);
2078 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2083 for (i = 0; i < ret; i++)
2084 btrfs_drop_and_free_fs_root(fs_info, gang[i]);
2088 static void btrfs_init_scrub(struct btrfs_fs_info *fs_info)
2090 mutex_init(&fs_info->scrub_lock);
2091 atomic_set(&fs_info->scrubs_running, 0);
2092 atomic_set(&fs_info->scrub_pause_req, 0);
2093 atomic_set(&fs_info->scrubs_paused, 0);
2094 atomic_set(&fs_info->scrub_cancel_req, 0);
2095 init_waitqueue_head(&fs_info->scrub_pause_wait);
2096 refcount_set(&fs_info->scrub_workers_refcnt, 0);
2099 static void btrfs_init_balance(struct btrfs_fs_info *fs_info)
2101 spin_lock_init(&fs_info->balance_lock);
2102 mutex_init(&fs_info->balance_mutex);
2103 atomic_set(&fs_info->balance_pause_req, 0);
2104 atomic_set(&fs_info->balance_cancel_req, 0);
2105 fs_info->balance_ctl = NULL;
2106 init_waitqueue_head(&fs_info->balance_wait_q);
2109 static void btrfs_init_btree_inode(struct btrfs_fs_info *fs_info)
2111 struct inode *inode = fs_info->btree_inode;
2113 inode->i_ino = BTRFS_BTREE_INODE_OBJECTID;
2114 set_nlink(inode, 1);
2116 * we set the i_size on the btree inode to the max possible int.
2117 * the real end of the address space is determined by all of
2118 * the devices in the system
2120 inode->i_size = OFFSET_MAX;
2121 inode->i_mapping->a_ops = &btree_aops;
2123 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
2124 extent_io_tree_init(fs_info, &BTRFS_I(inode)->io_tree,
2125 IO_TREE_BTREE_INODE_IO, inode);
2126 BTRFS_I(inode)->io_tree.track_uptodate = false;
2127 extent_map_tree_init(&BTRFS_I(inode)->extent_tree);
2129 BTRFS_I(inode)->root = btrfs_grab_root(fs_info->tree_root);
2130 memset(&BTRFS_I(inode)->location, 0, sizeof(struct btrfs_key));
2131 set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags);
2132 btrfs_insert_inode_hash(inode);
2135 static void btrfs_init_dev_replace_locks(struct btrfs_fs_info *fs_info)
2137 mutex_init(&fs_info->dev_replace.lock_finishing_cancel_unmount);
2138 init_rwsem(&fs_info->dev_replace.rwsem);
2139 init_waitqueue_head(&fs_info->dev_replace.replace_wait);
2142 static void btrfs_init_qgroup(struct btrfs_fs_info *fs_info)
2144 spin_lock_init(&fs_info->qgroup_lock);
2145 mutex_init(&fs_info->qgroup_ioctl_lock);
2146 fs_info->qgroup_tree = RB_ROOT;
2147 INIT_LIST_HEAD(&fs_info->dirty_qgroups);
2148 fs_info->qgroup_seq = 1;
2149 fs_info->qgroup_ulist = NULL;
2150 fs_info->qgroup_rescan_running = false;
2151 mutex_init(&fs_info->qgroup_rescan_lock);
2154 static int btrfs_init_workqueues(struct btrfs_fs_info *fs_info,
2155 struct btrfs_fs_devices *fs_devices)
2157 u32 max_active = fs_info->thread_pool_size;
2158 unsigned int flags = WQ_MEM_RECLAIM | WQ_FREEZABLE | WQ_UNBOUND;
2161 btrfs_alloc_workqueue(fs_info, "worker",
2162 flags | WQ_HIGHPRI, max_active, 16);
2164 fs_info->delalloc_workers =
2165 btrfs_alloc_workqueue(fs_info, "delalloc",
2166 flags, max_active, 2);
2168 fs_info->flush_workers =
2169 btrfs_alloc_workqueue(fs_info, "flush_delalloc",
2170 flags, max_active, 0);
2172 fs_info->caching_workers =
2173 btrfs_alloc_workqueue(fs_info, "cache", flags, max_active, 0);
2175 fs_info->fixup_workers =
2176 btrfs_alloc_workqueue(fs_info, "fixup", flags, 1, 0);
2179 * endios are largely parallel and should have a very
2182 fs_info->endio_workers =
2183 btrfs_alloc_workqueue(fs_info, "endio", flags, max_active, 4);
2184 fs_info->endio_meta_workers =
2185 btrfs_alloc_workqueue(fs_info, "endio-meta", flags,
2187 fs_info->endio_meta_write_workers =
2188 btrfs_alloc_workqueue(fs_info, "endio-meta-write", flags,
2190 fs_info->endio_raid56_workers =
2191 btrfs_alloc_workqueue(fs_info, "endio-raid56", flags,
2193 fs_info->rmw_workers =
2194 btrfs_alloc_workqueue(fs_info, "rmw", flags, max_active, 2);
2195 fs_info->endio_write_workers =
2196 btrfs_alloc_workqueue(fs_info, "endio-write", flags,
2198 fs_info->endio_freespace_worker =
2199 btrfs_alloc_workqueue(fs_info, "freespace-write", flags,
2201 fs_info->delayed_workers =
2202 btrfs_alloc_workqueue(fs_info, "delayed-meta", flags,
2204 fs_info->readahead_workers =
2205 btrfs_alloc_workqueue(fs_info, "readahead", flags,
2207 fs_info->qgroup_rescan_workers =
2208 btrfs_alloc_workqueue(fs_info, "qgroup-rescan", flags, 1, 0);
2209 fs_info->discard_ctl.discard_workers =
2210 alloc_workqueue("btrfs_discard", WQ_UNBOUND | WQ_FREEZABLE, 1);
2212 if (!(fs_info->workers && fs_info->delalloc_workers &&
2213 fs_info->flush_workers &&
2214 fs_info->endio_workers && fs_info->endio_meta_workers &&
2215 fs_info->endio_meta_write_workers &&
2216 fs_info->endio_write_workers && fs_info->endio_raid56_workers &&
2217 fs_info->endio_freespace_worker && fs_info->rmw_workers &&
2218 fs_info->caching_workers && fs_info->readahead_workers &&
2219 fs_info->fixup_workers && fs_info->delayed_workers &&
2220 fs_info->qgroup_rescan_workers &&
2221 fs_info->discard_ctl.discard_workers)) {
2228 static int btrfs_init_csum_hash(struct btrfs_fs_info *fs_info, u16 csum_type)
2230 struct crypto_shash *csum_shash;
2231 const char *csum_driver = btrfs_super_csum_driver(csum_type);
2233 csum_shash = crypto_alloc_shash(csum_driver, 0, 0);
2235 if (IS_ERR(csum_shash)) {
2236 btrfs_err(fs_info, "error allocating %s hash for checksum",
2238 return PTR_ERR(csum_shash);
2241 fs_info->csum_shash = csum_shash;
2246 static int btrfs_replay_log(struct btrfs_fs_info *fs_info,
2247 struct btrfs_fs_devices *fs_devices)
2250 struct btrfs_root *log_tree_root;
2251 struct btrfs_super_block *disk_super = fs_info->super_copy;
2252 u64 bytenr = btrfs_super_log_root(disk_super);
2253 int level = btrfs_super_log_root_level(disk_super);
2255 if (fs_devices->rw_devices == 0) {
2256 btrfs_warn(fs_info, "log replay required on RO media");
2260 log_tree_root = btrfs_alloc_root(fs_info, BTRFS_TREE_LOG_OBJECTID,
2265 log_tree_root->node = read_tree_block(fs_info, bytenr,
2266 fs_info->generation + 1,
2268 if (IS_ERR(log_tree_root->node)) {
2269 btrfs_warn(fs_info, "failed to read log tree");
2270 ret = PTR_ERR(log_tree_root->node);
2271 log_tree_root->node = NULL;
2272 btrfs_put_root(log_tree_root);
2274 } else if (!extent_buffer_uptodate(log_tree_root->node)) {
2275 btrfs_err(fs_info, "failed to read log tree");
2276 btrfs_put_root(log_tree_root);
2279 /* returns with log_tree_root freed on success */
2280 ret = btrfs_recover_log_trees(log_tree_root);
2282 btrfs_handle_fs_error(fs_info, ret,
2283 "Failed to recover log tree");
2284 btrfs_put_root(log_tree_root);
2288 if (sb_rdonly(fs_info->sb)) {
2289 ret = btrfs_commit_super(fs_info);
2297 static int btrfs_read_roots(struct btrfs_fs_info *fs_info)
2299 struct btrfs_root *tree_root = fs_info->tree_root;
2300 struct btrfs_root *root;
2301 struct btrfs_key location;
2304 BUG_ON(!fs_info->tree_root);
2306 location.objectid = BTRFS_EXTENT_TREE_OBJECTID;
2307 location.type = BTRFS_ROOT_ITEM_KEY;
2308 location.offset = 0;
2310 root = btrfs_read_tree_root(tree_root, &location);
2312 if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) {
2313 ret = PTR_ERR(root);
2317 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2318 fs_info->extent_root = root;
2321 location.objectid = BTRFS_DEV_TREE_OBJECTID;
2322 root = btrfs_read_tree_root(tree_root, &location);
2324 if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) {
2325 ret = PTR_ERR(root);
2329 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2330 fs_info->dev_root = root;
2331 btrfs_init_devices_late(fs_info);
2334 /* If IGNOREDATACSUMS is set don't bother reading the csum root. */
2335 if (!btrfs_test_opt(fs_info, IGNOREDATACSUMS)) {
2336 location.objectid = BTRFS_CSUM_TREE_OBJECTID;
2337 root = btrfs_read_tree_root(tree_root, &location);
2339 if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) {
2340 ret = PTR_ERR(root);
2344 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2345 fs_info->csum_root = root;
2350 * This tree can share blocks with some other fs tree during relocation
2351 * and we need a proper setup by btrfs_get_fs_root
2353 root = btrfs_get_fs_root(tree_root->fs_info,
2354 BTRFS_DATA_RELOC_TREE_OBJECTID, true);
2356 if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) {
2357 ret = PTR_ERR(root);
2361 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2362 fs_info->data_reloc_root = root;
2365 location.objectid = BTRFS_QUOTA_TREE_OBJECTID;
2366 root = btrfs_read_tree_root(tree_root, &location);
2367 if (!IS_ERR(root)) {
2368 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2369 set_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags);
2370 fs_info->quota_root = root;
2373 location.objectid = BTRFS_UUID_TREE_OBJECTID;
2374 root = btrfs_read_tree_root(tree_root, &location);
2376 if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) {
2377 ret = PTR_ERR(root);
2382 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2383 fs_info->uuid_root = root;
2386 if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
2387 location.objectid = BTRFS_FREE_SPACE_TREE_OBJECTID;
2388 root = btrfs_read_tree_root(tree_root, &location);
2390 if (!btrfs_test_opt(fs_info, IGNOREBADROOTS)) {
2391 ret = PTR_ERR(root);
2395 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2396 fs_info->free_space_root = root;
2402 btrfs_warn(fs_info, "failed to read root (objectid=%llu): %d",
2403 location.objectid, ret);
2408 * Real super block validation
2409 * NOTE: super csum type and incompat features will not be checked here.
2411 * @sb: super block to check
2412 * @mirror_num: the super block number to check its bytenr:
2413 * 0 the primary (1st) sb
2414 * 1, 2 2nd and 3rd backup copy
2415 * -1 skip bytenr check
2417 static int validate_super(struct btrfs_fs_info *fs_info,
2418 struct btrfs_super_block *sb, int mirror_num)
2420 u64 nodesize = btrfs_super_nodesize(sb);
2421 u64 sectorsize = btrfs_super_sectorsize(sb);
2424 if (btrfs_super_magic(sb) != BTRFS_MAGIC) {
2425 btrfs_err(fs_info, "no valid FS found");
2428 if (btrfs_super_flags(sb) & ~BTRFS_SUPER_FLAG_SUPP) {
2429 btrfs_err(fs_info, "unrecognized or unsupported super flag: %llu",
2430 btrfs_super_flags(sb) & ~BTRFS_SUPER_FLAG_SUPP);
2433 if (btrfs_super_root_level(sb) >= BTRFS_MAX_LEVEL) {
2434 btrfs_err(fs_info, "tree_root level too big: %d >= %d",
2435 btrfs_super_root_level(sb), BTRFS_MAX_LEVEL);
2438 if (btrfs_super_chunk_root_level(sb) >= BTRFS_MAX_LEVEL) {
2439 btrfs_err(fs_info, "chunk_root level too big: %d >= %d",
2440 btrfs_super_chunk_root_level(sb), BTRFS_MAX_LEVEL);
2443 if (btrfs_super_log_root_level(sb) >= BTRFS_MAX_LEVEL) {
2444 btrfs_err(fs_info, "log_root level too big: %d >= %d",
2445 btrfs_super_log_root_level(sb), BTRFS_MAX_LEVEL);
2450 * Check sectorsize and nodesize first, other check will need it.
2451 * Check all possible sectorsize(4K, 8K, 16K, 32K, 64K) here.
2453 if (!is_power_of_2(sectorsize) || sectorsize < 4096 ||
2454 sectorsize > BTRFS_MAX_METADATA_BLOCKSIZE) {
2455 btrfs_err(fs_info, "invalid sectorsize %llu", sectorsize);
2458 /* Only PAGE SIZE is supported yet */
2459 if (sectorsize != PAGE_SIZE) {
2461 "sectorsize %llu not supported yet, only support %lu",
2462 sectorsize, PAGE_SIZE);
2465 if (!is_power_of_2(nodesize) || nodesize < sectorsize ||
2466 nodesize > BTRFS_MAX_METADATA_BLOCKSIZE) {
2467 btrfs_err(fs_info, "invalid nodesize %llu", nodesize);
2470 if (nodesize != le32_to_cpu(sb->__unused_leafsize)) {
2471 btrfs_err(fs_info, "invalid leafsize %u, should be %llu",
2472 le32_to_cpu(sb->__unused_leafsize), nodesize);
2476 /* Root alignment check */
2477 if (!IS_ALIGNED(btrfs_super_root(sb), sectorsize)) {
2478 btrfs_warn(fs_info, "tree_root block unaligned: %llu",
2479 btrfs_super_root(sb));
2482 if (!IS_ALIGNED(btrfs_super_chunk_root(sb), sectorsize)) {
2483 btrfs_warn(fs_info, "chunk_root block unaligned: %llu",
2484 btrfs_super_chunk_root(sb));
2487 if (!IS_ALIGNED(btrfs_super_log_root(sb), sectorsize)) {
2488 btrfs_warn(fs_info, "log_root block unaligned: %llu",
2489 btrfs_super_log_root(sb));
2493 if (memcmp(fs_info->fs_devices->metadata_uuid, sb->dev_item.fsid,
2494 BTRFS_FSID_SIZE) != 0) {
2496 "dev_item UUID does not match metadata fsid: %pU != %pU",
2497 fs_info->fs_devices->metadata_uuid, sb->dev_item.fsid);
2502 * Hint to catch really bogus numbers, bitflips or so, more exact checks are
2505 if (btrfs_super_bytes_used(sb) < 6 * btrfs_super_nodesize(sb)) {
2506 btrfs_err(fs_info, "bytes_used is too small %llu",
2507 btrfs_super_bytes_used(sb));
2510 if (!is_power_of_2(btrfs_super_stripesize(sb))) {
2511 btrfs_err(fs_info, "invalid stripesize %u",
2512 btrfs_super_stripesize(sb));
2515 if (btrfs_super_num_devices(sb) > (1UL << 31))
2516 btrfs_warn(fs_info, "suspicious number of devices: %llu",
2517 btrfs_super_num_devices(sb));
2518 if (btrfs_super_num_devices(sb) == 0) {
2519 btrfs_err(fs_info, "number of devices is 0");
2523 if (mirror_num >= 0 &&
2524 btrfs_super_bytenr(sb) != btrfs_sb_offset(mirror_num)) {
2525 btrfs_err(fs_info, "super offset mismatch %llu != %u",
2526 btrfs_super_bytenr(sb), BTRFS_SUPER_INFO_OFFSET);
2531 * Obvious sys_chunk_array corruptions, it must hold at least one key
2534 if (btrfs_super_sys_array_size(sb) > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) {
2535 btrfs_err(fs_info, "system chunk array too big %u > %u",
2536 btrfs_super_sys_array_size(sb),
2537 BTRFS_SYSTEM_CHUNK_ARRAY_SIZE);
2540 if (btrfs_super_sys_array_size(sb) < sizeof(struct btrfs_disk_key)
2541 + sizeof(struct btrfs_chunk)) {
2542 btrfs_err(fs_info, "system chunk array too small %u < %zu",
2543 btrfs_super_sys_array_size(sb),
2544 sizeof(struct btrfs_disk_key)
2545 + sizeof(struct btrfs_chunk));
2550 * The generation is a global counter, we'll trust it more than the others
2551 * but it's still possible that it's the one that's wrong.
2553 if (btrfs_super_generation(sb) < btrfs_super_chunk_root_generation(sb))
2555 "suspicious: generation < chunk_root_generation: %llu < %llu",
2556 btrfs_super_generation(sb),
2557 btrfs_super_chunk_root_generation(sb));
2558 if (btrfs_super_generation(sb) < btrfs_super_cache_generation(sb)
2559 && btrfs_super_cache_generation(sb) != (u64)-1)
2561 "suspicious: generation < cache_generation: %llu < %llu",
2562 btrfs_super_generation(sb),
2563 btrfs_super_cache_generation(sb));
2569 * Validation of super block at mount time.
2570 * Some checks already done early at mount time, like csum type and incompat
2571 * flags will be skipped.
2573 static int btrfs_validate_mount_super(struct btrfs_fs_info *fs_info)
2575 return validate_super(fs_info, fs_info->super_copy, 0);
2579 * Validation of super block at write time.
2580 * Some checks like bytenr check will be skipped as their values will be
2582 * Extra checks like csum type and incompat flags will be done here.
2584 static int btrfs_validate_write_super(struct btrfs_fs_info *fs_info,
2585 struct btrfs_super_block *sb)
2589 ret = validate_super(fs_info, sb, -1);
2592 if (!btrfs_supported_super_csum(btrfs_super_csum_type(sb))) {
2594 btrfs_err(fs_info, "invalid csum type, has %u want %u",
2595 btrfs_super_csum_type(sb), BTRFS_CSUM_TYPE_CRC32);
2598 if (btrfs_super_incompat_flags(sb) & ~BTRFS_FEATURE_INCOMPAT_SUPP) {
2601 "invalid incompat flags, has 0x%llx valid mask 0x%llx",
2602 btrfs_super_incompat_flags(sb),
2603 (unsigned long long)BTRFS_FEATURE_INCOMPAT_SUPP);
2609 "super block corruption detected before writing it to disk");
2613 static int __cold init_tree_roots(struct btrfs_fs_info *fs_info)
2615 int backup_index = find_newest_super_backup(fs_info);
2616 struct btrfs_super_block *sb = fs_info->super_copy;
2617 struct btrfs_root *tree_root = fs_info->tree_root;
2618 bool handle_error = false;
2622 for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) {
2627 if (!IS_ERR(tree_root->node))
2628 free_extent_buffer(tree_root->node);
2629 tree_root->node = NULL;
2631 if (!btrfs_test_opt(fs_info, USEBACKUPROOT))
2634 free_root_pointers(fs_info, 0);
2637 * Don't use the log in recovery mode, it won't be
2640 btrfs_set_super_log_root(sb, 0);
2642 /* We can't trust the free space cache either */
2643 btrfs_set_opt(fs_info->mount_opt, CLEAR_CACHE);
2645 ret = read_backup_root(fs_info, i);
2650 generation = btrfs_super_generation(sb);
2651 level = btrfs_super_root_level(sb);
2652 tree_root->node = read_tree_block(fs_info, btrfs_super_root(sb),
2653 generation, level, NULL);
2654 if (IS_ERR(tree_root->node)) {
2655 handle_error = true;
2656 ret = PTR_ERR(tree_root->node);
2657 tree_root->node = NULL;
2658 btrfs_warn(fs_info, "couldn't read tree root");
2661 } else if (!extent_buffer_uptodate(tree_root->node)) {
2662 handle_error = true;
2664 btrfs_warn(fs_info, "error while reading tree root");
2668 btrfs_set_root_node(&tree_root->root_item, tree_root->node);
2669 tree_root->commit_root = btrfs_root_node(tree_root);
2670 btrfs_set_root_refs(&tree_root->root_item, 1);
2673 * No need to hold btrfs_root::objectid_mutex since the fs
2674 * hasn't been fully initialised and we are the only user
2676 ret = btrfs_find_highest_objectid(tree_root,
2677 &tree_root->highest_objectid);
2679 handle_error = true;
2683 ASSERT(tree_root->highest_objectid <= BTRFS_LAST_FREE_OBJECTID);
2685 ret = btrfs_read_roots(fs_info);
2687 handle_error = true;
2691 /* All successful */
2692 fs_info->generation = generation;
2693 fs_info->last_trans_committed = generation;
2695 /* Always begin writing backup roots after the one being used */
2696 if (backup_index < 0) {
2697 fs_info->backup_root_index = 0;
2699 fs_info->backup_root_index = backup_index + 1;
2700 fs_info->backup_root_index %= BTRFS_NUM_BACKUP_ROOTS;
2708 void btrfs_init_fs_info(struct btrfs_fs_info *fs_info)
2710 INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC);
2711 INIT_RADIX_TREE(&fs_info->buffer_radix, GFP_ATOMIC);
2712 INIT_LIST_HEAD(&fs_info->trans_list);
2713 INIT_LIST_HEAD(&fs_info->dead_roots);
2714 INIT_LIST_HEAD(&fs_info->delayed_iputs);
2715 INIT_LIST_HEAD(&fs_info->delalloc_roots);
2716 INIT_LIST_HEAD(&fs_info->caching_block_groups);
2717 spin_lock_init(&fs_info->delalloc_root_lock);
2718 spin_lock_init(&fs_info->trans_lock);
2719 spin_lock_init(&fs_info->fs_roots_radix_lock);
2720 spin_lock_init(&fs_info->delayed_iput_lock);
2721 spin_lock_init(&fs_info->defrag_inodes_lock);
2722 spin_lock_init(&fs_info->super_lock);
2723 spin_lock_init(&fs_info->buffer_lock);
2724 spin_lock_init(&fs_info->unused_bgs_lock);
2725 rwlock_init(&fs_info->tree_mod_log_lock);
2726 mutex_init(&fs_info->unused_bg_unpin_mutex);
2727 mutex_init(&fs_info->delete_unused_bgs_mutex);
2728 mutex_init(&fs_info->reloc_mutex);
2729 mutex_init(&fs_info->delalloc_root_mutex);
2730 seqlock_init(&fs_info->profiles_lock);
2732 INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
2733 INIT_LIST_HEAD(&fs_info->space_info);
2734 INIT_LIST_HEAD(&fs_info->tree_mod_seq_list);
2735 INIT_LIST_HEAD(&fs_info->unused_bgs);
2736 #ifdef CONFIG_BTRFS_DEBUG
2737 INIT_LIST_HEAD(&fs_info->allocated_roots);
2738 INIT_LIST_HEAD(&fs_info->allocated_ebs);
2739 spin_lock_init(&fs_info->eb_leak_lock);
2741 extent_map_tree_init(&fs_info->mapping_tree);
2742 btrfs_init_block_rsv(&fs_info->global_block_rsv,
2743 BTRFS_BLOCK_RSV_GLOBAL);
2744 btrfs_init_block_rsv(&fs_info->trans_block_rsv, BTRFS_BLOCK_RSV_TRANS);
2745 btrfs_init_block_rsv(&fs_info->chunk_block_rsv, BTRFS_BLOCK_RSV_CHUNK);
2746 btrfs_init_block_rsv(&fs_info->empty_block_rsv, BTRFS_BLOCK_RSV_EMPTY);
2747 btrfs_init_block_rsv(&fs_info->delayed_block_rsv,
2748 BTRFS_BLOCK_RSV_DELOPS);
2749 btrfs_init_block_rsv(&fs_info->delayed_refs_rsv,
2750 BTRFS_BLOCK_RSV_DELREFS);
2752 atomic_set(&fs_info->async_delalloc_pages, 0);
2753 atomic_set(&fs_info->defrag_running, 0);
2754 atomic_set(&fs_info->reada_works_cnt, 0);
2755 atomic_set(&fs_info->nr_delayed_iputs, 0);
2756 atomic64_set(&fs_info->tree_mod_seq, 0);
2757 fs_info->max_inline = BTRFS_DEFAULT_MAX_INLINE;
2758 fs_info->metadata_ratio = 0;
2759 fs_info->defrag_inodes = RB_ROOT;
2760 atomic64_set(&fs_info->free_chunk_space, 0);
2761 fs_info->tree_mod_log = RB_ROOT;
2762 fs_info->commit_interval = BTRFS_DEFAULT_COMMIT_INTERVAL;
2763 fs_info->avg_delayed_ref_runtime = NSEC_PER_SEC >> 6; /* div by 64 */
2764 /* readahead state */
2765 INIT_RADIX_TREE(&fs_info->reada_tree, GFP_NOFS & ~__GFP_DIRECT_RECLAIM);
2766 spin_lock_init(&fs_info->reada_lock);
2767 btrfs_init_ref_verify(fs_info);
2769 fs_info->thread_pool_size = min_t(unsigned long,
2770 num_online_cpus() + 2, 8);
2772 INIT_LIST_HEAD(&fs_info->ordered_roots);
2773 spin_lock_init(&fs_info->ordered_root_lock);
2775 btrfs_init_scrub(fs_info);
2776 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2777 fs_info->check_integrity_print_mask = 0;
2779 btrfs_init_balance(fs_info);
2780 btrfs_init_async_reclaim_work(fs_info);
2782 spin_lock_init(&fs_info->block_group_cache_lock);
2783 fs_info->block_group_cache_tree = RB_ROOT;
2784 fs_info->first_logical_byte = (u64)-1;
2786 extent_io_tree_init(fs_info, &fs_info->excluded_extents,
2787 IO_TREE_FS_EXCLUDED_EXTENTS, NULL);
2788 set_bit(BTRFS_FS_BARRIER, &fs_info->flags);
2790 mutex_init(&fs_info->ordered_operations_mutex);
2791 mutex_init(&fs_info->tree_log_mutex);
2792 mutex_init(&fs_info->chunk_mutex);
2793 mutex_init(&fs_info->transaction_kthread_mutex);
2794 mutex_init(&fs_info->cleaner_mutex);
2795 mutex_init(&fs_info->ro_block_group_mutex);
2796 init_rwsem(&fs_info->commit_root_sem);
2797 init_rwsem(&fs_info->cleanup_work_sem);
2798 init_rwsem(&fs_info->subvol_sem);
2799 sema_init(&fs_info->uuid_tree_rescan_sem, 1);
2801 btrfs_init_dev_replace_locks(fs_info);
2802 btrfs_init_qgroup(fs_info);
2803 btrfs_discard_init(fs_info);
2805 btrfs_init_free_cluster(&fs_info->meta_alloc_cluster);
2806 btrfs_init_free_cluster(&fs_info->data_alloc_cluster);
2808 init_waitqueue_head(&fs_info->transaction_throttle);
2809 init_waitqueue_head(&fs_info->transaction_wait);
2810 init_waitqueue_head(&fs_info->transaction_blocked_wait);
2811 init_waitqueue_head(&fs_info->async_submit_wait);
2812 init_waitqueue_head(&fs_info->delayed_iputs_wait);
2814 /* Usable values until the real ones are cached from the superblock */
2815 fs_info->nodesize = 4096;
2816 fs_info->sectorsize = 4096;
2817 fs_info->sectorsize_bits = ilog2(4096);
2818 fs_info->stripesize = 4096;
2820 spin_lock_init(&fs_info->swapfile_pins_lock);
2821 fs_info->swapfile_pins = RB_ROOT;
2823 fs_info->send_in_progress = 0;
2826 static int init_mount_fs_info(struct btrfs_fs_info *fs_info, struct super_block *sb)
2831 sb->s_blocksize = BTRFS_BDEV_BLOCKSIZE;
2832 sb->s_blocksize_bits = blksize_bits(BTRFS_BDEV_BLOCKSIZE);
2834 ret = percpu_counter_init(&fs_info->dio_bytes, 0, GFP_KERNEL);
2838 ret = percpu_counter_init(&fs_info->dirty_metadata_bytes, 0, GFP_KERNEL);
2842 fs_info->dirty_metadata_batch = PAGE_SIZE *
2843 (1 + ilog2(nr_cpu_ids));
2845 ret = percpu_counter_init(&fs_info->delalloc_bytes, 0, GFP_KERNEL);
2849 ret = percpu_counter_init(&fs_info->dev_replace.bio_counter, 0,
2854 fs_info->delayed_root = kmalloc(sizeof(struct btrfs_delayed_root),
2856 if (!fs_info->delayed_root)
2858 btrfs_init_delayed_root(fs_info->delayed_root);
2860 return btrfs_alloc_stripe_hash_table(fs_info);
2863 static int btrfs_uuid_rescan_kthread(void *data)
2865 struct btrfs_fs_info *fs_info = (struct btrfs_fs_info *)data;
2869 * 1st step is to iterate through the existing UUID tree and
2870 * to delete all entries that contain outdated data.
2871 * 2nd step is to add all missing entries to the UUID tree.
2873 ret = btrfs_uuid_tree_iterate(fs_info);
2876 btrfs_warn(fs_info, "iterating uuid_tree failed %d",
2878 up(&fs_info->uuid_tree_rescan_sem);
2881 return btrfs_uuid_scan_kthread(data);
2884 static int btrfs_check_uuid_tree(struct btrfs_fs_info *fs_info)
2886 struct task_struct *task;
2888 down(&fs_info->uuid_tree_rescan_sem);
2889 task = kthread_run(btrfs_uuid_rescan_kthread, fs_info, "btrfs-uuid");
2891 /* fs_info->update_uuid_tree_gen remains 0 in all error case */
2892 btrfs_warn(fs_info, "failed to start uuid_rescan task");
2893 up(&fs_info->uuid_tree_rescan_sem);
2894 return PTR_ERR(task);
2900 int __cold open_ctree(struct super_block *sb, struct btrfs_fs_devices *fs_devices,
2909 struct btrfs_super_block *disk_super;
2910 struct btrfs_fs_info *fs_info = btrfs_sb(sb);
2911 struct btrfs_root *tree_root;
2912 struct btrfs_root *chunk_root;
2915 int clear_free_space_tree = 0;
2918 ret = init_mount_fs_info(fs_info, sb);
2924 /* These need to be init'ed before we start creating inodes and such. */
2925 tree_root = btrfs_alloc_root(fs_info, BTRFS_ROOT_TREE_OBJECTID,
2927 fs_info->tree_root = tree_root;
2928 chunk_root = btrfs_alloc_root(fs_info, BTRFS_CHUNK_TREE_OBJECTID,
2930 fs_info->chunk_root = chunk_root;
2931 if (!tree_root || !chunk_root) {
2936 fs_info->btree_inode = new_inode(sb);
2937 if (!fs_info->btree_inode) {
2941 mapping_set_gfp_mask(fs_info->btree_inode->i_mapping, GFP_NOFS);
2942 btrfs_init_btree_inode(fs_info);
2944 invalidate_bdev(fs_devices->latest_bdev);
2947 * Read super block and check the signature bytes only
2949 disk_super = btrfs_read_dev_super(fs_devices->latest_bdev);
2950 if (IS_ERR(disk_super)) {
2951 err = PTR_ERR(disk_super);
2956 * Verify the type first, if that or the checksum value are
2957 * corrupted, we'll find out
2959 csum_type = btrfs_super_csum_type(disk_super);
2960 if (!btrfs_supported_super_csum(csum_type)) {
2961 btrfs_err(fs_info, "unsupported checksum algorithm: %u",
2964 btrfs_release_disk_super(disk_super);
2968 ret = btrfs_init_csum_hash(fs_info, csum_type);
2971 btrfs_release_disk_super(disk_super);
2976 * We want to check superblock checksum, the type is stored inside.
2977 * Pass the whole disk block of size BTRFS_SUPER_INFO_SIZE (4k).
2979 if (btrfs_check_super_csum(fs_info, (u8 *)disk_super)) {
2980 btrfs_err(fs_info, "superblock checksum mismatch");
2982 btrfs_release_disk_super(disk_super);
2987 * super_copy is zeroed at allocation time and we never touch the
2988 * following bytes up to INFO_SIZE, the checksum is calculated from
2989 * the whole block of INFO_SIZE
2991 memcpy(fs_info->super_copy, disk_super, sizeof(*fs_info->super_copy));
2992 btrfs_release_disk_super(disk_super);
2994 disk_super = fs_info->super_copy;
2996 ASSERT(!memcmp(fs_info->fs_devices->fsid, fs_info->super_copy->fsid,
2999 if (btrfs_fs_incompat(fs_info, METADATA_UUID)) {
3000 ASSERT(!memcmp(fs_info->fs_devices->metadata_uuid,
3001 fs_info->super_copy->metadata_uuid,
3005 features = btrfs_super_flags(disk_super);
3006 if (features & BTRFS_SUPER_FLAG_CHANGING_FSID_V2) {
3007 features &= ~BTRFS_SUPER_FLAG_CHANGING_FSID_V2;
3008 btrfs_set_super_flags(disk_super, features);
3010 "found metadata UUID change in progress flag, clearing");
3013 memcpy(fs_info->super_for_commit, fs_info->super_copy,
3014 sizeof(*fs_info->super_for_commit));
3016 ret = btrfs_validate_mount_super(fs_info);
3018 btrfs_err(fs_info, "superblock contains fatal errors");
3023 if (!btrfs_super_root(disk_super))
3026 /* check FS state, whether FS is broken. */
3027 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_ERROR)
3028 set_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state);
3031 * In the long term, we'll store the compression type in the super
3032 * block, and it'll be used for per file compression control.
3034 fs_info->compress_type = BTRFS_COMPRESS_ZLIB;
3036 ret = btrfs_parse_options(fs_info, options, sb->s_flags);
3042 features = btrfs_super_incompat_flags(disk_super) &
3043 ~BTRFS_FEATURE_INCOMPAT_SUPP;
3046 "cannot mount because of unsupported optional features (%llx)",
3052 features = btrfs_super_incompat_flags(disk_super);
3053 features |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF;
3054 if (fs_info->compress_type == BTRFS_COMPRESS_LZO)
3055 features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO;
3056 else if (fs_info->compress_type == BTRFS_COMPRESS_ZSTD)
3057 features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_ZSTD;
3059 if (features & BTRFS_FEATURE_INCOMPAT_SKINNY_METADATA)
3060 btrfs_info(fs_info, "has skinny extents");
3063 * flag our filesystem as having big metadata blocks if
3064 * they are bigger than the page size
3066 if (btrfs_super_nodesize(disk_super) > PAGE_SIZE) {
3067 if (!(features & BTRFS_FEATURE_INCOMPAT_BIG_METADATA))
3069 "flagging fs with big metadata feature");
3070 features |= BTRFS_FEATURE_INCOMPAT_BIG_METADATA;
3073 nodesize = btrfs_super_nodesize(disk_super);
3074 sectorsize = btrfs_super_sectorsize(disk_super);
3075 stripesize = sectorsize;
3076 fs_info->dirty_metadata_batch = nodesize * (1 + ilog2(nr_cpu_ids));
3077 fs_info->delalloc_batch = sectorsize * 512 * (1 + ilog2(nr_cpu_ids));
3079 /* Cache block sizes */
3080 fs_info->nodesize = nodesize;
3081 fs_info->sectorsize = sectorsize;
3082 fs_info->sectorsize_bits = ilog2(sectorsize);
3083 fs_info->csum_size = btrfs_super_csum_size(disk_super);
3084 fs_info->csums_per_leaf = BTRFS_MAX_ITEM_SIZE(fs_info) / fs_info->csum_size;
3085 fs_info->stripesize = stripesize;
3088 * mixed block groups end up with duplicate but slightly offset
3089 * extent buffers for the same range. It leads to corruptions
3091 if ((features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) &&
3092 (sectorsize != nodesize)) {
3094 "unequal nodesize/sectorsize (%u != %u) are not allowed for mixed block groups",
3095 nodesize, sectorsize);
3100 * Needn't use the lock because there is no other task which will
3103 btrfs_set_super_incompat_flags(disk_super, features);
3105 features = btrfs_super_compat_ro_flags(disk_super) &
3106 ~BTRFS_FEATURE_COMPAT_RO_SUPP;
3107 if (!sb_rdonly(sb) && features) {
3109 "cannot mount read-write because of unsupported optional features (%llx)",
3115 ret = btrfs_init_workqueues(fs_info, fs_devices);
3118 goto fail_sb_buffer;
3121 sb->s_bdi->ra_pages *= btrfs_super_num_devices(disk_super);
3122 sb->s_bdi->ra_pages = max(sb->s_bdi->ra_pages, SZ_4M / PAGE_SIZE);
3124 sb->s_blocksize = sectorsize;
3125 sb->s_blocksize_bits = blksize_bits(sectorsize);
3126 memcpy(&sb->s_uuid, fs_info->fs_devices->fsid, BTRFS_FSID_SIZE);
3128 mutex_lock(&fs_info->chunk_mutex);
3129 ret = btrfs_read_sys_array(fs_info);
3130 mutex_unlock(&fs_info->chunk_mutex);
3132 btrfs_err(fs_info, "failed to read the system array: %d", ret);
3133 goto fail_sb_buffer;
3136 generation = btrfs_super_chunk_root_generation(disk_super);
3137 level = btrfs_super_chunk_root_level(disk_super);
3139 chunk_root->node = read_tree_block(fs_info,
3140 btrfs_super_chunk_root(disk_super),
3141 generation, level, NULL);
3142 if (IS_ERR(chunk_root->node) ||
3143 !extent_buffer_uptodate(chunk_root->node)) {
3144 btrfs_err(fs_info, "failed to read chunk root");
3145 if (!IS_ERR(chunk_root->node))
3146 free_extent_buffer(chunk_root->node);
3147 chunk_root->node = NULL;
3148 goto fail_tree_roots;
3150 btrfs_set_root_node(&chunk_root->root_item, chunk_root->node);
3151 chunk_root->commit_root = btrfs_root_node(chunk_root);
3153 read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
3154 offsetof(struct btrfs_header, chunk_tree_uuid),
3157 ret = btrfs_read_chunk_tree(fs_info);
3159 btrfs_err(fs_info, "failed to read chunk tree: %d", ret);
3160 goto fail_tree_roots;
3164 * Keep the devid that is marked to be the target device for the
3165 * device replace procedure
3167 btrfs_free_extra_devids(fs_devices, 0);
3169 if (!fs_devices->latest_bdev) {
3170 btrfs_err(fs_info, "failed to read devices");
3171 goto fail_tree_roots;
3174 ret = init_tree_roots(fs_info);
3176 goto fail_tree_roots;
3179 * If we have a uuid root and we're not being told to rescan we need to
3180 * check the generation here so we can set the
3181 * BTRFS_FS_UPDATE_UUID_TREE_GEN bit. Otherwise we could commit the
3182 * transaction during a balance or the log replay without updating the
3183 * uuid generation, and then if we crash we would rescan the uuid tree,
3184 * even though it was perfectly fine.
3186 if (fs_info->uuid_root && !btrfs_test_opt(fs_info, RESCAN_UUID_TREE) &&
3187 fs_info->generation == btrfs_super_uuid_tree_generation(disk_super))
3188 set_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN, &fs_info->flags);
3190 ret = btrfs_verify_dev_extents(fs_info);
3193 "failed to verify dev extents against chunks: %d",
3195 goto fail_block_groups;
3197 ret = btrfs_recover_balance(fs_info);
3199 btrfs_err(fs_info, "failed to recover balance: %d", ret);
3200 goto fail_block_groups;
3203 ret = btrfs_init_dev_stats(fs_info);
3205 btrfs_err(fs_info, "failed to init dev_stats: %d", ret);
3206 goto fail_block_groups;
3209 ret = btrfs_init_dev_replace(fs_info);
3211 btrfs_err(fs_info, "failed to init dev_replace: %d", ret);
3212 goto fail_block_groups;
3215 btrfs_free_extra_devids(fs_devices, 1);
3217 ret = btrfs_sysfs_add_fsid(fs_devices);
3219 btrfs_err(fs_info, "failed to init sysfs fsid interface: %d",
3221 goto fail_block_groups;
3224 ret = btrfs_sysfs_add_mounted(fs_info);
3226 btrfs_err(fs_info, "failed to init sysfs interface: %d", ret);
3227 goto fail_fsdev_sysfs;
3230 ret = btrfs_init_space_info(fs_info);
3232 btrfs_err(fs_info, "failed to initialize space info: %d", ret);
3236 ret = btrfs_read_block_groups(fs_info);
3238 btrfs_err(fs_info, "failed to read block groups: %d", ret);
3242 if (!sb_rdonly(sb) && !btrfs_check_rw_degradable(fs_info, NULL)) {
3244 "writable mount is not allowed due to too many missing devices");
3248 fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root,
3250 if (IS_ERR(fs_info->cleaner_kthread))
3253 fs_info->transaction_kthread = kthread_run(transaction_kthread,
3255 "btrfs-transaction");
3256 if (IS_ERR(fs_info->transaction_kthread))
3259 if (!btrfs_test_opt(fs_info, NOSSD) &&
3260 !fs_info->fs_devices->rotating) {
3261 btrfs_set_and_info(fs_info, SSD, "enabling ssd optimizations");
3265 * Mount does not set all options immediately, we can do it now and do
3266 * not have to wait for transaction commit
3268 btrfs_apply_pending_changes(fs_info);
3270 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3271 if (btrfs_test_opt(fs_info, CHECK_INTEGRITY)) {
3272 ret = btrfsic_mount(fs_info, fs_devices,
3273 btrfs_test_opt(fs_info,
3274 CHECK_INTEGRITY_INCLUDING_EXTENT_DATA) ?
3276 fs_info->check_integrity_print_mask);
3279 "failed to initialize integrity check module: %d",
3283 ret = btrfs_read_qgroup_config(fs_info);
3285 goto fail_trans_kthread;
3287 if (btrfs_build_ref_tree(fs_info))
3288 btrfs_err(fs_info, "couldn't build ref tree");
3290 /* do not make disk changes in broken FS or nologreplay is given */
3291 if (btrfs_super_log_root(disk_super) != 0 &&
3292 !btrfs_test_opt(fs_info, NOLOGREPLAY)) {
3293 btrfs_info(fs_info, "start tree-log replay");
3294 ret = btrfs_replay_log(fs_info, fs_devices);
3301 ret = btrfs_find_orphan_roots(fs_info);
3305 if (!sb_rdonly(sb)) {
3306 ret = btrfs_cleanup_fs_roots(fs_info);
3310 mutex_lock(&fs_info->cleaner_mutex);
3311 ret = btrfs_recover_relocation(tree_root);
3312 mutex_unlock(&fs_info->cleaner_mutex);
3314 btrfs_warn(fs_info, "failed to recover relocation: %d",
3321 fs_info->fs_root = btrfs_get_fs_root(fs_info, BTRFS_FS_TREE_OBJECTID, true);
3322 if (IS_ERR(fs_info->fs_root)) {
3323 err = PTR_ERR(fs_info->fs_root);
3324 btrfs_warn(fs_info, "failed to read fs tree: %d", err);
3325 fs_info->fs_root = NULL;
3332 if (btrfs_test_opt(fs_info, CLEAR_CACHE) &&
3333 btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
3334 clear_free_space_tree = 1;
3335 } else if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE) &&
3336 !btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE_VALID)) {
3337 btrfs_warn(fs_info, "free space tree is invalid");
3338 clear_free_space_tree = 1;
3341 if (clear_free_space_tree) {
3342 btrfs_info(fs_info, "clearing free space tree");
3343 ret = btrfs_clear_free_space_tree(fs_info);
3346 "failed to clear free space tree: %d", ret);
3347 close_ctree(fs_info);
3352 if (btrfs_test_opt(fs_info, FREE_SPACE_TREE) &&
3353 !btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
3354 btrfs_info(fs_info, "creating free space tree");
3355 ret = btrfs_create_free_space_tree(fs_info);
3358 "failed to create free space tree: %d", ret);
3359 close_ctree(fs_info);
3364 down_read(&fs_info->cleanup_work_sem);
3365 if ((ret = btrfs_orphan_cleanup(fs_info->fs_root)) ||
3366 (ret = btrfs_orphan_cleanup(fs_info->tree_root))) {
3367 up_read(&fs_info->cleanup_work_sem);
3368 close_ctree(fs_info);
3371 up_read(&fs_info->cleanup_work_sem);
3373 ret = btrfs_resume_balance_async(fs_info);
3375 btrfs_warn(fs_info, "failed to resume balance: %d", ret);
3376 close_ctree(fs_info);
3380 ret = btrfs_resume_dev_replace_async(fs_info);
3382 btrfs_warn(fs_info, "failed to resume device replace: %d", ret);
3383 close_ctree(fs_info);
3387 btrfs_qgroup_rescan_resume(fs_info);
3388 btrfs_discard_resume(fs_info);
3390 if (!fs_info->uuid_root) {
3391 btrfs_info(fs_info, "creating UUID tree");
3392 ret = btrfs_create_uuid_tree(fs_info);
3395 "failed to create the UUID tree: %d", ret);
3396 close_ctree(fs_info);
3399 } else if (btrfs_test_opt(fs_info, RESCAN_UUID_TREE) ||
3400 fs_info->generation !=
3401 btrfs_super_uuid_tree_generation(disk_super)) {
3402 btrfs_info(fs_info, "checking UUID tree");
3403 ret = btrfs_check_uuid_tree(fs_info);
3406 "failed to check the UUID tree: %d", ret);
3407 close_ctree(fs_info);
3411 set_bit(BTRFS_FS_OPEN, &fs_info->flags);
3416 btrfs_free_qgroup_config(fs_info);
3418 kthread_stop(fs_info->transaction_kthread);
3419 btrfs_cleanup_transaction(fs_info);
3420 btrfs_free_fs_roots(fs_info);
3422 kthread_stop(fs_info->cleaner_kthread);
3425 * make sure we're done with the btree inode before we stop our
3428 filemap_write_and_wait(fs_info->btree_inode->i_mapping);
3431 btrfs_sysfs_remove_mounted(fs_info);
3434 btrfs_sysfs_remove_fsid(fs_info->fs_devices);
3437 btrfs_put_block_group_cache(fs_info);
3440 if (fs_info->data_reloc_root)
3441 btrfs_drop_and_free_fs_root(fs_info, fs_info->data_reloc_root);
3442 free_root_pointers(fs_info, true);
3443 invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
3446 btrfs_stop_all_workers(fs_info);
3447 btrfs_free_block_groups(fs_info);
3449 btrfs_mapping_tree_free(&fs_info->mapping_tree);
3451 iput(fs_info->btree_inode);
3453 btrfs_close_devices(fs_info->fs_devices);
3456 ALLOW_ERROR_INJECTION(open_ctree, ERRNO);
3458 static void btrfs_end_super_write(struct bio *bio)
3460 struct btrfs_device *device = bio->bi_private;
3461 struct bio_vec *bvec;
3462 struct bvec_iter_all iter_all;
3465 bio_for_each_segment_all(bvec, bio, iter_all) {
3466 page = bvec->bv_page;
3468 if (bio->bi_status) {
3469 btrfs_warn_rl_in_rcu(device->fs_info,
3470 "lost page write due to IO error on %s (%d)",
3471 rcu_str_deref(device->name),
3472 blk_status_to_errno(bio->bi_status));
3473 ClearPageUptodate(page);
3475 btrfs_dev_stat_inc_and_print(device,
3476 BTRFS_DEV_STAT_WRITE_ERRS);
3478 SetPageUptodate(page);
3488 struct btrfs_super_block *btrfs_read_dev_one_super(struct block_device *bdev,
3491 struct btrfs_super_block *super;
3494 struct address_space *mapping = bdev->bd_inode->i_mapping;
3496 bytenr = btrfs_sb_offset(copy_num);
3497 if (bytenr + BTRFS_SUPER_INFO_SIZE >= i_size_read(bdev->bd_inode))
3498 return ERR_PTR(-EINVAL);
3500 page = read_cache_page_gfp(mapping, bytenr >> PAGE_SHIFT, GFP_NOFS);
3502 return ERR_CAST(page);
3504 super = page_address(page);
3505 if (btrfs_super_magic(super) != BTRFS_MAGIC) {
3506 btrfs_release_disk_super(super);
3507 return ERR_PTR(-ENODATA);
3510 if (btrfs_super_bytenr(super) != bytenr) {
3511 btrfs_release_disk_super(super);
3512 return ERR_PTR(-EINVAL);
3519 struct btrfs_super_block *btrfs_read_dev_super(struct block_device *bdev)
3521 struct btrfs_super_block *super, *latest = NULL;
3525 /* we would like to check all the supers, but that would make
3526 * a btrfs mount succeed after a mkfs from a different FS.
3527 * So, we need to add a special mount option to scan for
3528 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
3530 for (i = 0; i < 1; i++) {
3531 super = btrfs_read_dev_one_super(bdev, i);
3535 if (!latest || btrfs_super_generation(super) > transid) {
3537 btrfs_release_disk_super(super);
3540 transid = btrfs_super_generation(super);
3548 * Write superblock @sb to the @device. Do not wait for completion, all the
3549 * pages we use for writing are locked.
3551 * Write @max_mirrors copies of the superblock, where 0 means default that fit
3552 * the expected device size at commit time. Note that max_mirrors must be
3553 * same for write and wait phases.
3555 * Return number of errors when page is not found or submission fails.
3557 static int write_dev_supers(struct btrfs_device *device,
3558 struct btrfs_super_block *sb, int max_mirrors)
3560 struct btrfs_fs_info *fs_info = device->fs_info;
3561 struct address_space *mapping = device->bdev->bd_inode->i_mapping;
3562 SHASH_DESC_ON_STACK(shash, fs_info->csum_shash);
3567 if (max_mirrors == 0)
3568 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
3570 shash->tfm = fs_info->csum_shash;
3572 for (i = 0; i < max_mirrors; i++) {
3575 struct btrfs_super_block *disk_super;
3577 bytenr = btrfs_sb_offset(i);
3578 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
3579 device->commit_total_bytes)
3582 btrfs_set_super_bytenr(sb, bytenr);
3584 crypto_shash_digest(shash, (const char *)sb + BTRFS_CSUM_SIZE,
3585 BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE,
3588 page = find_or_create_page(mapping, bytenr >> PAGE_SHIFT,
3591 btrfs_err(device->fs_info,
3592 "couldn't get super block page for bytenr %llu",
3598 /* Bump the refcount for wait_dev_supers() */
3601 disk_super = page_address(page);
3602 memcpy(disk_super, sb, BTRFS_SUPER_INFO_SIZE);
3605 * Directly use bios here instead of relying on the page cache
3606 * to do I/O, so we don't lose the ability to do integrity
3609 bio = bio_alloc(GFP_NOFS, 1);
3610 bio_set_dev(bio, device->bdev);
3611 bio->bi_iter.bi_sector = bytenr >> SECTOR_SHIFT;
3612 bio->bi_private = device;
3613 bio->bi_end_io = btrfs_end_super_write;
3614 __bio_add_page(bio, page, BTRFS_SUPER_INFO_SIZE,
3615 offset_in_page(bytenr));
3618 * We FUA only the first super block. The others we allow to
3619 * go down lazy and there's a short window where the on-disk
3620 * copies might still contain the older version.
3622 bio->bi_opf = REQ_OP_WRITE | REQ_SYNC | REQ_META | REQ_PRIO;
3623 if (i == 0 && !btrfs_test_opt(device->fs_info, NOBARRIER))
3624 bio->bi_opf |= REQ_FUA;
3626 btrfsic_submit_bio(bio);
3628 return errors < i ? 0 : -1;
3632 * Wait for write completion of superblocks done by write_dev_supers,
3633 * @max_mirrors same for write and wait phases.
3635 * Return number of errors when page is not found or not marked up to
3638 static int wait_dev_supers(struct btrfs_device *device, int max_mirrors)
3642 bool primary_failed = false;
3645 if (max_mirrors == 0)
3646 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
3648 for (i = 0; i < max_mirrors; i++) {
3651 bytenr = btrfs_sb_offset(i);
3652 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
3653 device->commit_total_bytes)
3656 page = find_get_page(device->bdev->bd_inode->i_mapping,
3657 bytenr >> PAGE_SHIFT);
3661 primary_failed = true;
3664 /* Page is submitted locked and unlocked once the IO completes */
3665 wait_on_page_locked(page);
3666 if (PageError(page)) {
3669 primary_failed = true;
3672 /* Drop our reference */
3675 /* Drop the reference from the writing run */
3679 /* log error, force error return */
3680 if (primary_failed) {
3681 btrfs_err(device->fs_info, "error writing primary super block to device %llu",
3686 return errors < i ? 0 : -1;
3690 * endio for the write_dev_flush, this will wake anyone waiting
3691 * for the barrier when it is done
3693 static void btrfs_end_empty_barrier(struct bio *bio)
3695 complete(bio->bi_private);
3699 * Submit a flush request to the device if it supports it. Error handling is
3700 * done in the waiting counterpart.
3702 static void write_dev_flush(struct btrfs_device *device)
3704 struct request_queue *q = bdev_get_queue(device->bdev);
3705 struct bio *bio = device->flush_bio;
3707 if (!test_bit(QUEUE_FLAG_WC, &q->queue_flags))
3711 bio->bi_end_io = btrfs_end_empty_barrier;
3712 bio_set_dev(bio, device->bdev);
3713 bio->bi_opf = REQ_OP_WRITE | REQ_SYNC | REQ_PREFLUSH;
3714 init_completion(&device->flush_wait);
3715 bio->bi_private = &device->flush_wait;
3717 btrfsic_submit_bio(bio);
3718 set_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state);
3722 * If the flush bio has been submitted by write_dev_flush, wait for it.
3724 static blk_status_t wait_dev_flush(struct btrfs_device *device)
3726 struct bio *bio = device->flush_bio;
3728 if (!test_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state))
3731 clear_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state);
3732 wait_for_completion_io(&device->flush_wait);
3734 return bio->bi_status;
3737 static int check_barrier_error(struct btrfs_fs_info *fs_info)
3739 if (!btrfs_check_rw_degradable(fs_info, NULL))
3745 * send an empty flush down to each device in parallel,
3746 * then wait for them
3748 static int barrier_all_devices(struct btrfs_fs_info *info)
3750 struct list_head *head;
3751 struct btrfs_device *dev;
3752 int errors_wait = 0;
3755 lockdep_assert_held(&info->fs_devices->device_list_mutex);
3756 /* send down all the barriers */
3757 head = &info->fs_devices->devices;
3758 list_for_each_entry(dev, head, dev_list) {
3759 if (test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state))
3763 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
3764 !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
3767 write_dev_flush(dev);
3768 dev->last_flush_error = BLK_STS_OK;
3771 /* wait for all the barriers */
3772 list_for_each_entry(dev, head, dev_list) {
3773 if (test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state))
3779 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
3780 !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
3783 ret = wait_dev_flush(dev);
3785 dev->last_flush_error = ret;
3786 btrfs_dev_stat_inc_and_print(dev,
3787 BTRFS_DEV_STAT_FLUSH_ERRS);
3794 * At some point we need the status of all disks
3795 * to arrive at the volume status. So error checking
3796 * is being pushed to a separate loop.
3798 return check_barrier_error(info);
3803 int btrfs_get_num_tolerated_disk_barrier_failures(u64 flags)
3806 int min_tolerated = INT_MAX;
3808 if ((flags & BTRFS_BLOCK_GROUP_PROFILE_MASK) == 0 ||
3809 (flags & BTRFS_AVAIL_ALLOC_BIT_SINGLE))
3810 min_tolerated = min_t(int, min_tolerated,
3811 btrfs_raid_array[BTRFS_RAID_SINGLE].
3812 tolerated_failures);
3814 for (raid_type = 0; raid_type < BTRFS_NR_RAID_TYPES; raid_type++) {
3815 if (raid_type == BTRFS_RAID_SINGLE)
3817 if (!(flags & btrfs_raid_array[raid_type].bg_flag))
3819 min_tolerated = min_t(int, min_tolerated,
3820 btrfs_raid_array[raid_type].
3821 tolerated_failures);
3824 if (min_tolerated == INT_MAX) {
3825 pr_warn("BTRFS: unknown raid flag: %llu", flags);
3829 return min_tolerated;
3832 int write_all_supers(struct btrfs_fs_info *fs_info, int max_mirrors)
3834 struct list_head *head;
3835 struct btrfs_device *dev;
3836 struct btrfs_super_block *sb;
3837 struct btrfs_dev_item *dev_item;
3841 int total_errors = 0;
3844 do_barriers = !btrfs_test_opt(fs_info, NOBARRIER);
3847 * max_mirrors == 0 indicates we're from commit_transaction,
3848 * not from fsync where the tree roots in fs_info have not
3849 * been consistent on disk.
3851 if (max_mirrors == 0)
3852 backup_super_roots(fs_info);
3854 sb = fs_info->super_for_commit;
3855 dev_item = &sb->dev_item;
3857 mutex_lock(&fs_info->fs_devices->device_list_mutex);
3858 head = &fs_info->fs_devices->devices;
3859 max_errors = btrfs_super_num_devices(fs_info->super_copy) - 1;
3862 ret = barrier_all_devices(fs_info);
3865 &fs_info->fs_devices->device_list_mutex);
3866 btrfs_handle_fs_error(fs_info, ret,
3867 "errors while submitting device barriers.");
3872 list_for_each_entry(dev, head, dev_list) {
3877 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
3878 !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
3881 btrfs_set_stack_device_generation(dev_item, 0);
3882 btrfs_set_stack_device_type(dev_item, dev->type);
3883 btrfs_set_stack_device_id(dev_item, dev->devid);
3884 btrfs_set_stack_device_total_bytes(dev_item,
3885 dev->commit_total_bytes);
3886 btrfs_set_stack_device_bytes_used(dev_item,
3887 dev->commit_bytes_used);
3888 btrfs_set_stack_device_io_align(dev_item, dev->io_align);
3889 btrfs_set_stack_device_io_width(dev_item, dev->io_width);
3890 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
3891 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
3892 memcpy(dev_item->fsid, dev->fs_devices->metadata_uuid,
3895 flags = btrfs_super_flags(sb);
3896 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
3898 ret = btrfs_validate_write_super(fs_info, sb);
3900 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
3901 btrfs_handle_fs_error(fs_info, -EUCLEAN,
3902 "unexpected superblock corruption detected");
3906 ret = write_dev_supers(dev, sb, max_mirrors);
3910 if (total_errors > max_errors) {
3911 btrfs_err(fs_info, "%d errors while writing supers",
3913 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
3915 /* FUA is masked off if unsupported and can't be the reason */
3916 btrfs_handle_fs_error(fs_info, -EIO,
3917 "%d errors while writing supers",
3923 list_for_each_entry(dev, head, dev_list) {
3926 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &dev->dev_state) ||
3927 !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))
3930 ret = wait_dev_supers(dev, max_mirrors);
3934 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
3935 if (total_errors > max_errors) {
3936 btrfs_handle_fs_error(fs_info, -EIO,
3937 "%d errors while writing supers",
3944 /* Drop a fs root from the radix tree and free it. */
3945 void btrfs_drop_and_free_fs_root(struct btrfs_fs_info *fs_info,
3946 struct btrfs_root *root)
3948 bool drop_ref = false;
3950 spin_lock(&fs_info->fs_roots_radix_lock);
3951 radix_tree_delete(&fs_info->fs_roots_radix,
3952 (unsigned long)root->root_key.objectid);
3953 if (test_and_clear_bit(BTRFS_ROOT_IN_RADIX, &root->state))
3955 spin_unlock(&fs_info->fs_roots_radix_lock);
3957 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
3958 ASSERT(root->log_root == NULL);
3959 if (root->reloc_root) {
3960 btrfs_put_root(root->reloc_root);
3961 root->reloc_root = NULL;
3965 if (root->free_ino_pinned)
3966 __btrfs_remove_free_space_cache(root->free_ino_pinned);
3967 if (root->free_ino_ctl)
3968 __btrfs_remove_free_space_cache(root->free_ino_ctl);
3969 if (root->ino_cache_inode) {
3970 iput(root->ino_cache_inode);
3971 root->ino_cache_inode = NULL;
3974 btrfs_put_root(root);
3977 int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info)
3979 u64 root_objectid = 0;
3980 struct btrfs_root *gang[8];
3983 unsigned int ret = 0;
3986 spin_lock(&fs_info->fs_roots_radix_lock);
3987 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
3988 (void **)gang, root_objectid,
3991 spin_unlock(&fs_info->fs_roots_radix_lock);
3994 root_objectid = gang[ret - 1]->root_key.objectid + 1;
3996 for (i = 0; i < ret; i++) {
3997 /* Avoid to grab roots in dead_roots */
3998 if (btrfs_root_refs(&gang[i]->root_item) == 0) {
4002 /* grab all the search result for later use */
4003 gang[i] = btrfs_grab_root(gang[i]);
4005 spin_unlock(&fs_info->fs_roots_radix_lock);
4007 for (i = 0; i < ret; i++) {
4010 root_objectid = gang[i]->root_key.objectid;
4011 err = btrfs_orphan_cleanup(gang[i]);
4014 btrfs_put_root(gang[i]);
4019 /* release the uncleaned roots due to error */
4020 for (; i < ret; i++) {
4022 btrfs_put_root(gang[i]);
4027 int btrfs_commit_super(struct btrfs_fs_info *fs_info)
4029 struct btrfs_root *root = fs_info->tree_root;
4030 struct btrfs_trans_handle *trans;
4032 mutex_lock(&fs_info->cleaner_mutex);
4033 btrfs_run_delayed_iputs(fs_info);
4034 mutex_unlock(&fs_info->cleaner_mutex);
4035 wake_up_process(fs_info->cleaner_kthread);
4037 /* wait until ongoing cleanup work done */
4038 down_write(&fs_info->cleanup_work_sem);
4039 up_write(&fs_info->cleanup_work_sem);
4041 trans = btrfs_join_transaction(root);
4043 return PTR_ERR(trans);
4044 return btrfs_commit_transaction(trans);
4047 void __cold close_ctree(struct btrfs_fs_info *fs_info)
4051 set_bit(BTRFS_FS_CLOSING_START, &fs_info->flags);
4053 * We don't want the cleaner to start new transactions, add more delayed
4054 * iputs, etc. while we're closing. We can't use kthread_stop() yet
4055 * because that frees the task_struct, and the transaction kthread might
4056 * still try to wake up the cleaner.
4058 kthread_park(fs_info->cleaner_kthread);
4060 /* wait for the qgroup rescan worker to stop */
4061 btrfs_qgroup_wait_for_completion(fs_info, false);
4063 /* wait for the uuid_scan task to finish */
4064 down(&fs_info->uuid_tree_rescan_sem);
4065 /* avoid complains from lockdep et al., set sem back to initial state */
4066 up(&fs_info->uuid_tree_rescan_sem);
4068 /* pause restriper - we want to resume on mount */
4069 btrfs_pause_balance(fs_info);
4071 btrfs_dev_replace_suspend_for_unmount(fs_info);
4073 btrfs_scrub_cancel(fs_info);
4075 /* wait for any defraggers to finish */
4076 wait_event(fs_info->transaction_wait,
4077 (atomic_read(&fs_info->defrag_running) == 0));
4079 /* clear out the rbtree of defraggable inodes */
4080 btrfs_cleanup_defrag_inodes(fs_info);
4082 cancel_work_sync(&fs_info->async_reclaim_work);
4083 cancel_work_sync(&fs_info->async_data_reclaim_work);
4085 /* Cancel or finish ongoing discard work */
4086 btrfs_discard_cleanup(fs_info);
4088 if (!sb_rdonly(fs_info->sb)) {
4090 * The cleaner kthread is stopped, so do one final pass over
4091 * unused block groups.
4093 btrfs_delete_unused_bgs(fs_info);
4096 * There might be existing delayed inode workers still running
4097 * and holding an empty delayed inode item. We must wait for
4098 * them to complete first because they can create a transaction.
4099 * This happens when someone calls btrfs_balance_delayed_items()
4100 * and then a transaction commit runs the same delayed nodes
4101 * before any delayed worker has done something with the nodes.
4102 * We must wait for any worker here and not at transaction
4103 * commit time since that could cause a deadlock.
4104 * This is a very rare case.
4106 btrfs_flush_workqueue(fs_info->delayed_workers);
4108 ret = btrfs_commit_super(fs_info);
4110 btrfs_err(fs_info, "commit super ret %d", ret);
4113 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state) ||
4114 test_bit(BTRFS_FS_STATE_TRANS_ABORTED, &fs_info->fs_state))
4115 btrfs_error_commit_super(fs_info);
4117 kthread_stop(fs_info->transaction_kthread);
4118 kthread_stop(fs_info->cleaner_kthread);
4120 ASSERT(list_empty(&fs_info->delayed_iputs));
4121 set_bit(BTRFS_FS_CLOSING_DONE, &fs_info->flags);
4123 if (btrfs_check_quota_leak(fs_info)) {
4124 WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG));
4125 btrfs_err(fs_info, "qgroup reserved space leaked");
4128 btrfs_free_qgroup_config(fs_info);
4129 ASSERT(list_empty(&fs_info->delalloc_roots));
4131 if (percpu_counter_sum(&fs_info->delalloc_bytes)) {
4132 btrfs_info(fs_info, "at unmount delalloc count %lld",
4133 percpu_counter_sum(&fs_info->delalloc_bytes));
4136 if (percpu_counter_sum(&fs_info->dio_bytes))
4137 btrfs_info(fs_info, "at unmount dio bytes count %lld",
4138 percpu_counter_sum(&fs_info->dio_bytes));
4140 btrfs_sysfs_remove_mounted(fs_info);
4141 btrfs_sysfs_remove_fsid(fs_info->fs_devices);
4143 btrfs_put_block_group_cache(fs_info);
4146 * we must make sure there is not any read request to
4147 * submit after we stopping all workers.
4149 invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
4150 btrfs_stop_all_workers(fs_info);
4152 clear_bit(BTRFS_FS_OPEN, &fs_info->flags);
4153 free_root_pointers(fs_info, true);
4154 btrfs_free_fs_roots(fs_info);
4157 * We must free the block groups after dropping the fs_roots as we could
4158 * have had an IO error and have left over tree log blocks that aren't
4159 * cleaned up until the fs roots are freed. This makes the block group
4160 * accounting appear to be wrong because there's pending reserved bytes,
4161 * so make sure we do the block group cleanup afterwards.
4163 btrfs_free_block_groups(fs_info);
4165 iput(fs_info->btree_inode);
4167 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
4168 if (btrfs_test_opt(fs_info, CHECK_INTEGRITY))
4169 btrfsic_unmount(fs_info->fs_devices);
4172 btrfs_mapping_tree_free(&fs_info->mapping_tree);
4173 btrfs_close_devices(fs_info->fs_devices);
4176 int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid,
4180 struct inode *btree_inode = buf->pages[0]->mapping->host;
4182 ret = extent_buffer_uptodate(buf);
4186 ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf,
4187 parent_transid, atomic);
4193 void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
4195 struct btrfs_fs_info *fs_info;
4196 struct btrfs_root *root;
4197 u64 transid = btrfs_header_generation(buf);
4200 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
4202 * This is a fast path so only do this check if we have sanity tests
4203 * enabled. Normal people shouldn't be using unmapped buffers as dirty
4204 * outside of the sanity tests.
4206 if (unlikely(test_bit(EXTENT_BUFFER_UNMAPPED, &buf->bflags)))
4209 root = BTRFS_I(buf->pages[0]->mapping->host)->root;
4210 fs_info = root->fs_info;
4211 btrfs_assert_tree_locked(buf);
4212 if (transid != fs_info->generation)
4213 WARN(1, KERN_CRIT "btrfs transid mismatch buffer %llu, found %llu running %llu\n",
4214 buf->start, transid, fs_info->generation);
4215 was_dirty = set_extent_buffer_dirty(buf);
4217 percpu_counter_add_batch(&fs_info->dirty_metadata_bytes,
4219 fs_info->dirty_metadata_batch);
4220 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
4222 * Since btrfs_mark_buffer_dirty() can be called with item pointer set
4223 * but item data not updated.
4224 * So here we should only check item pointers, not item data.
4226 if (btrfs_header_level(buf) == 0 &&
4227 btrfs_check_leaf_relaxed(buf)) {
4228 btrfs_print_leaf(buf);
4234 static void __btrfs_btree_balance_dirty(struct btrfs_fs_info *fs_info,
4238 * looks as though older kernels can get into trouble with
4239 * this code, they end up stuck in balance_dirty_pages forever
4243 if (current->flags & PF_MEMALLOC)
4247 btrfs_balance_delayed_items(fs_info);
4249 ret = __percpu_counter_compare(&fs_info->dirty_metadata_bytes,
4250 BTRFS_DIRTY_METADATA_THRESH,
4251 fs_info->dirty_metadata_batch);
4253 balance_dirty_pages_ratelimited(fs_info->btree_inode->i_mapping);
4257 void btrfs_btree_balance_dirty(struct btrfs_fs_info *fs_info)
4259 __btrfs_btree_balance_dirty(fs_info, 1);
4262 void btrfs_btree_balance_dirty_nodelay(struct btrfs_fs_info *fs_info)
4264 __btrfs_btree_balance_dirty(fs_info, 0);
4267 int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid, int level,
4268 struct btrfs_key *first_key)
4270 return btree_read_extent_buffer_pages(buf, parent_transid,
4274 static void btrfs_error_commit_super(struct btrfs_fs_info *fs_info)
4276 /* cleanup FS via transaction */
4277 btrfs_cleanup_transaction(fs_info);
4279 mutex_lock(&fs_info->cleaner_mutex);
4280 btrfs_run_delayed_iputs(fs_info);
4281 mutex_unlock(&fs_info->cleaner_mutex);
4283 down_write(&fs_info->cleanup_work_sem);
4284 up_write(&fs_info->cleanup_work_sem);
4287 static void btrfs_drop_all_logs(struct btrfs_fs_info *fs_info)
4289 struct btrfs_root *gang[8];
4290 u64 root_objectid = 0;
4293 spin_lock(&fs_info->fs_roots_radix_lock);
4294 while ((ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
4295 (void **)gang, root_objectid,
4296 ARRAY_SIZE(gang))) != 0) {
4299 for (i = 0; i < ret; i++)
4300 gang[i] = btrfs_grab_root(gang[i]);
4301 spin_unlock(&fs_info->fs_roots_radix_lock);
4303 for (i = 0; i < ret; i++) {
4306 root_objectid = gang[i]->root_key.objectid;
4307 btrfs_free_log(NULL, gang[i]);
4308 btrfs_put_root(gang[i]);
4311 spin_lock(&fs_info->fs_roots_radix_lock);
4313 spin_unlock(&fs_info->fs_roots_radix_lock);
4314 btrfs_free_log_root_tree(NULL, fs_info);
4317 static void btrfs_destroy_ordered_extents(struct btrfs_root *root)
4319 struct btrfs_ordered_extent *ordered;
4321 spin_lock(&root->ordered_extent_lock);
4323 * This will just short circuit the ordered completion stuff which will
4324 * make sure the ordered extent gets properly cleaned up.
4326 list_for_each_entry(ordered, &root->ordered_extents,
4328 set_bit(BTRFS_ORDERED_IOERR, &ordered->flags);
4329 spin_unlock(&root->ordered_extent_lock);
4332 static void btrfs_destroy_all_ordered_extents(struct btrfs_fs_info *fs_info)
4334 struct btrfs_root *root;
4335 struct list_head splice;
4337 INIT_LIST_HEAD(&splice);
4339 spin_lock(&fs_info->ordered_root_lock);
4340 list_splice_init(&fs_info->ordered_roots, &splice);
4341 while (!list_empty(&splice)) {
4342 root = list_first_entry(&splice, struct btrfs_root,
4344 list_move_tail(&root->ordered_root,
4345 &fs_info->ordered_roots);
4347 spin_unlock(&fs_info->ordered_root_lock);
4348 btrfs_destroy_ordered_extents(root);
4351 spin_lock(&fs_info->ordered_root_lock);
4353 spin_unlock(&fs_info->ordered_root_lock);
4356 * We need this here because if we've been flipped read-only we won't
4357 * get sync() from the umount, so we need to make sure any ordered
4358 * extents that haven't had their dirty pages IO start writeout yet
4359 * actually get run and error out properly.
4361 btrfs_wait_ordered_roots(fs_info, U64_MAX, 0, (u64)-1);
4364 static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
4365 struct btrfs_fs_info *fs_info)
4367 struct rb_node *node;
4368 struct btrfs_delayed_ref_root *delayed_refs;
4369 struct btrfs_delayed_ref_node *ref;
4372 delayed_refs = &trans->delayed_refs;
4374 spin_lock(&delayed_refs->lock);
4375 if (atomic_read(&delayed_refs->num_entries) == 0) {
4376 spin_unlock(&delayed_refs->lock);
4377 btrfs_debug(fs_info, "delayed_refs has NO entry");
4381 while ((node = rb_first_cached(&delayed_refs->href_root)) != NULL) {
4382 struct btrfs_delayed_ref_head *head;
4384 bool pin_bytes = false;
4386 head = rb_entry(node, struct btrfs_delayed_ref_head,
4388 if (btrfs_delayed_ref_lock(delayed_refs, head))
4391 spin_lock(&head->lock);
4392 while ((n = rb_first_cached(&head->ref_tree)) != NULL) {
4393 ref = rb_entry(n, struct btrfs_delayed_ref_node,
4396 rb_erase_cached(&ref->ref_node, &head->ref_tree);
4397 RB_CLEAR_NODE(&ref->ref_node);
4398 if (!list_empty(&ref->add_list))
4399 list_del(&ref->add_list);
4400 atomic_dec(&delayed_refs->num_entries);
4401 btrfs_put_delayed_ref(ref);
4403 if (head->must_insert_reserved)
4405 btrfs_free_delayed_extent_op(head->extent_op);
4406 btrfs_delete_ref_head(delayed_refs, head);
4407 spin_unlock(&head->lock);
4408 spin_unlock(&delayed_refs->lock);
4409 mutex_unlock(&head->mutex);
4412 struct btrfs_block_group *cache;
4414 cache = btrfs_lookup_block_group(fs_info, head->bytenr);
4417 spin_lock(&cache->space_info->lock);
4418 spin_lock(&cache->lock);
4419 cache->pinned += head->num_bytes;
4420 btrfs_space_info_update_bytes_pinned(fs_info,
4421 cache->space_info, head->num_bytes);
4422 cache->reserved -= head->num_bytes;
4423 cache->space_info->bytes_reserved -= head->num_bytes;
4424 spin_unlock(&cache->lock);
4425 spin_unlock(&cache->space_info->lock);
4426 percpu_counter_add_batch(
4427 &cache->space_info->total_bytes_pinned,
4428 head->num_bytes, BTRFS_TOTAL_BYTES_PINNED_BATCH);
4430 btrfs_put_block_group(cache);
4432 btrfs_error_unpin_extent_range(fs_info, head->bytenr,
4433 head->bytenr + head->num_bytes - 1);
4435 btrfs_cleanup_ref_head_accounting(fs_info, delayed_refs, head);
4436 btrfs_put_delayed_ref_head(head);
4438 spin_lock(&delayed_refs->lock);
4440 btrfs_qgroup_destroy_extent_records(trans);
4442 spin_unlock(&delayed_refs->lock);
4447 static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root)
4449 struct btrfs_inode *btrfs_inode;
4450 struct list_head splice;
4452 INIT_LIST_HEAD(&splice);
4454 spin_lock(&root->delalloc_lock);
4455 list_splice_init(&root->delalloc_inodes, &splice);
4457 while (!list_empty(&splice)) {
4458 struct inode *inode = NULL;
4459 btrfs_inode = list_first_entry(&splice, struct btrfs_inode,
4461 __btrfs_del_delalloc_inode(root, btrfs_inode);
4462 spin_unlock(&root->delalloc_lock);
4465 * Make sure we get a live inode and that it'll not disappear
4468 inode = igrab(&btrfs_inode->vfs_inode);
4470 invalidate_inode_pages2(inode->i_mapping);
4473 spin_lock(&root->delalloc_lock);
4475 spin_unlock(&root->delalloc_lock);
4478 static void btrfs_destroy_all_delalloc_inodes(struct btrfs_fs_info *fs_info)
4480 struct btrfs_root *root;
4481 struct list_head splice;
4483 INIT_LIST_HEAD(&splice);
4485 spin_lock(&fs_info->delalloc_root_lock);
4486 list_splice_init(&fs_info->delalloc_roots, &splice);
4487 while (!list_empty(&splice)) {
4488 root = list_first_entry(&splice, struct btrfs_root,
4490 root = btrfs_grab_root(root);
4492 spin_unlock(&fs_info->delalloc_root_lock);
4494 btrfs_destroy_delalloc_inodes(root);
4495 btrfs_put_root(root);
4497 spin_lock(&fs_info->delalloc_root_lock);
4499 spin_unlock(&fs_info->delalloc_root_lock);
4502 static int btrfs_destroy_marked_extents(struct btrfs_fs_info *fs_info,
4503 struct extent_io_tree *dirty_pages,
4507 struct extent_buffer *eb;
4512 ret = find_first_extent_bit(dirty_pages, start, &start, &end,
4517 clear_extent_bits(dirty_pages, start, end, mark);
4518 while (start <= end) {
4519 eb = find_extent_buffer(fs_info, start);
4520 start += fs_info->nodesize;
4523 wait_on_extent_buffer_writeback(eb);
4525 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY,
4527 clear_extent_buffer_dirty(eb);
4528 free_extent_buffer_stale(eb);
4535 static int btrfs_destroy_pinned_extent(struct btrfs_fs_info *fs_info,
4536 struct extent_io_tree *unpin)
4543 struct extent_state *cached_state = NULL;
4546 * The btrfs_finish_extent_commit() may get the same range as
4547 * ours between find_first_extent_bit and clear_extent_dirty.
4548 * Hence, hold the unused_bg_unpin_mutex to avoid double unpin
4549 * the same extent range.
4551 mutex_lock(&fs_info->unused_bg_unpin_mutex);
4552 ret = find_first_extent_bit(unpin, 0, &start, &end,
4553 EXTENT_DIRTY, &cached_state);
4555 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
4559 clear_extent_dirty(unpin, start, end, &cached_state);
4560 free_extent_state(cached_state);
4561 btrfs_error_unpin_extent_range(fs_info, start, end);
4562 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
4569 static void btrfs_cleanup_bg_io(struct btrfs_block_group *cache)
4571 struct inode *inode;
4573 inode = cache->io_ctl.inode;
4575 invalidate_inode_pages2(inode->i_mapping);
4576 BTRFS_I(inode)->generation = 0;
4577 cache->io_ctl.inode = NULL;
4580 ASSERT(cache->io_ctl.pages == NULL);
4581 btrfs_put_block_group(cache);
4584 void btrfs_cleanup_dirty_bgs(struct btrfs_transaction *cur_trans,
4585 struct btrfs_fs_info *fs_info)
4587 struct btrfs_block_group *cache;
4589 spin_lock(&cur_trans->dirty_bgs_lock);
4590 while (!list_empty(&cur_trans->dirty_bgs)) {
4591 cache = list_first_entry(&cur_trans->dirty_bgs,
4592 struct btrfs_block_group,
4595 if (!list_empty(&cache->io_list)) {
4596 spin_unlock(&cur_trans->dirty_bgs_lock);
4597 list_del_init(&cache->io_list);
4598 btrfs_cleanup_bg_io(cache);
4599 spin_lock(&cur_trans->dirty_bgs_lock);
4602 list_del_init(&cache->dirty_list);
4603 spin_lock(&cache->lock);
4604 cache->disk_cache_state = BTRFS_DC_ERROR;
4605 spin_unlock(&cache->lock);
4607 spin_unlock(&cur_trans->dirty_bgs_lock);
4608 btrfs_put_block_group(cache);
4609 btrfs_delayed_refs_rsv_release(fs_info, 1);
4610 spin_lock(&cur_trans->dirty_bgs_lock);
4612 spin_unlock(&cur_trans->dirty_bgs_lock);
4615 * Refer to the definition of io_bgs member for details why it's safe
4616 * to use it without any locking
4618 while (!list_empty(&cur_trans->io_bgs)) {
4619 cache = list_first_entry(&cur_trans->io_bgs,
4620 struct btrfs_block_group,
4623 list_del_init(&cache->io_list);
4624 spin_lock(&cache->lock);
4625 cache->disk_cache_state = BTRFS_DC_ERROR;
4626 spin_unlock(&cache->lock);
4627 btrfs_cleanup_bg_io(cache);
4631 void btrfs_cleanup_one_transaction(struct btrfs_transaction *cur_trans,
4632 struct btrfs_fs_info *fs_info)
4634 struct btrfs_device *dev, *tmp;
4636 btrfs_cleanup_dirty_bgs(cur_trans, fs_info);
4637 ASSERT(list_empty(&cur_trans->dirty_bgs));
4638 ASSERT(list_empty(&cur_trans->io_bgs));
4640 list_for_each_entry_safe(dev, tmp, &cur_trans->dev_update_list,
4642 list_del_init(&dev->post_commit_list);
4645 btrfs_destroy_delayed_refs(cur_trans, fs_info);
4647 cur_trans->state = TRANS_STATE_COMMIT_START;
4648 wake_up(&fs_info->transaction_blocked_wait);
4650 cur_trans->state = TRANS_STATE_UNBLOCKED;
4651 wake_up(&fs_info->transaction_wait);
4653 btrfs_destroy_delayed_inodes(fs_info);
4655 btrfs_destroy_marked_extents(fs_info, &cur_trans->dirty_pages,
4657 btrfs_destroy_pinned_extent(fs_info, &cur_trans->pinned_extents);
4659 cur_trans->state =TRANS_STATE_COMPLETED;
4660 wake_up(&cur_trans->commit_wait);
4663 static int btrfs_cleanup_transaction(struct btrfs_fs_info *fs_info)
4665 struct btrfs_transaction *t;
4667 mutex_lock(&fs_info->transaction_kthread_mutex);
4669 spin_lock(&fs_info->trans_lock);
4670 while (!list_empty(&fs_info->trans_list)) {
4671 t = list_first_entry(&fs_info->trans_list,
4672 struct btrfs_transaction, list);
4673 if (t->state >= TRANS_STATE_COMMIT_START) {
4674 refcount_inc(&t->use_count);
4675 spin_unlock(&fs_info->trans_lock);
4676 btrfs_wait_for_commit(fs_info, t->transid);
4677 btrfs_put_transaction(t);
4678 spin_lock(&fs_info->trans_lock);
4681 if (t == fs_info->running_transaction) {
4682 t->state = TRANS_STATE_COMMIT_DOING;
4683 spin_unlock(&fs_info->trans_lock);
4685 * We wait for 0 num_writers since we don't hold a trans
4686 * handle open currently for this transaction.
4688 wait_event(t->writer_wait,
4689 atomic_read(&t->num_writers) == 0);
4691 spin_unlock(&fs_info->trans_lock);
4693 btrfs_cleanup_one_transaction(t, fs_info);
4695 spin_lock(&fs_info->trans_lock);
4696 if (t == fs_info->running_transaction)
4697 fs_info->running_transaction = NULL;
4698 list_del_init(&t->list);
4699 spin_unlock(&fs_info->trans_lock);
4701 btrfs_put_transaction(t);
4702 trace_btrfs_transaction_commit(fs_info->tree_root);
4703 spin_lock(&fs_info->trans_lock);
4705 spin_unlock(&fs_info->trans_lock);
4706 btrfs_destroy_all_ordered_extents(fs_info);
4707 btrfs_destroy_delayed_inodes(fs_info);
4708 btrfs_assert_delayed_root_empty(fs_info);
4709 btrfs_destroy_all_delalloc_inodes(fs_info);
4710 btrfs_drop_all_logs(fs_info);
4711 mutex_unlock(&fs_info->transaction_kthread_mutex);